Device performance detection method and device, intelligent agent, electronic device, and storage medium

Abstract

This invention provides a method, apparatus, intelligent agent, electronic device, and storage medium for equipment performance testing, belonging to the field of intelligent manufacturing technology. The method includes: obtaining the maximum cycle time of the production line where the target equipment is located; obtaining a candidate detection instruction set, which contains multiple independent atomic detection instructions, each instruction configured with a risk weight value, execution time, and execution correlation; selecting target atomic detection instructions from the candidate detection instruction set and arranging them to generate a target detection instruction sequence, with the optimization objective being to maximize the total risk weight value and the constraints being that the total execution time does not exceed the maximum cycle time and the execution correlation is satisfied; and sending the sequence to the target equipment. This invention achieves decoupling of the detection logic and the electronic control firmware by atomically decoupling the detection instructions and generating the detection sequence based on multi-constraint dynamic programming, supporting rapid iteration of the detection strategy, and maximizing quality risk coverage within a limited cycle time.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method, apparatus, intelligent agent, electronic device, and storage medium for testing equipment performance. Background Technology

[0002] Home appliances such as refrigerators and washing machines have many internal parts. Before they are assembled on the production line and come off the line, the working status of each major functional module needs to be tested, and the test must be completed within the cycle time specified by the production line.

[0003] To meet the above requirements, existing technologies typically employ statically fixed testing and control schemes. Specifically, engineers pre-select a set of high-risk testing items based on experience and write the corresponding operating procedures for each testing item into the equipment's electronic control firmware using a fixed, continuous logic hard-coded method. Before the production line goes offline, this performance testing mode is activated via button combinations or communication protocols. The electronic control firmware then runs each operating condition sequentially according to the pre-fixed order to complete the testing of each functional module.

[0004] However, the existing equipment performance testing schemes have significant shortcomings in practical applications. On the one hand, because the items to be tested are pre-coded into the electronic control firmware with fixed, continuous logic, the composition and execution order of the test items cannot be dynamically adjusted according to changes in actual quality risks during production line operation. This results in limited cycle time resources being uniformly consumed on experience-based pre-set test items, making it difficult to allocate testing resources to truly high-risk items. On the other hand, under the rigid constraint of production line cycle time, due to the complex coupling constraints such as execution dependencies and mutual exclusion relationships between test items, the existing fixed-sequence execution method is prone to resource misallocation, where cycle time is excessively occupied by low-value test items while high-value test items are forced to be abandoned due to insufficient time. This leads to a limitation on the overall defect detection capability. Summary of the Invention

[0005] This invention provides a method, apparatus, intelligent agent, electronic device, and storage medium for equipment performance testing. It addresses the shortcomings of existing production line offline testing items being fixed and preset, unable to be dynamically adjusted according to actual risks, and difficult to achieve optimal overall testing benefits under limited cycle time and complex execution constraints. The invention achieves risk weighting based on each independent atomic testing instruction. Under the constraints of cycle time limits and execution correlation, it dynamically plans and generates a target testing instruction sequence that maximizes the total risk weight value and issues it for execution. This allows limited cycle time resources to be precisely allocated to high-risk items, significantly improving the defect detection capability per unit testing time.

[0006] This invention provides a method for testing equipment performance, comprising: Obtain the maximum cycle time limit of the production line where the target equipment is located; Obtain a candidate detection instruction set, which contains multiple independent atomic detection instructions, each of which is configured with a risk weight value, execution time, and execution correlation. With the goal of maximizing the total risk weight value, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, multiple target atomic detection instructions are selected from the candidate detection instruction set and arranged in sequence to generate a target detection instruction sequence. The target detection instruction sequence is sent to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0007] According to a device performance testing method provided by the present invention, the step of selecting multiple target atomized detection instructions from the candidate detection instruction set and arranging them in sequence to generate a target detection instruction sequence includes: According to the risk weight values ​​from high to low, the atomic detection instructions in the candidate detection instruction set are sorted to obtain the sorted candidate detection instruction sequence. According to the sorted candidate detection instruction sequence, each of the atomic detection instructions is taken as the current instruction to be evaluated in turn, and it is determined whether the current instruction to be evaluated satisfies the constraint conditions. If the conditions are met, the current instruction to be evaluated is added to the selected instruction set as the target atomization detection instruction; if the conditions are not met, the current instruction to be evaluated is skipped. After determining all the atomic detection instructions in the sorted candidate detection instruction sequence, the target atomic detection instructions in the selected instruction set are arranged in the order of addition to generate the target detection instruction sequence.

[0008] According to a device performance testing method provided by the present invention, determining whether the current instruction to be evaluated satisfies the constraint conditions includes: Initialize the remaining available time to the maximum cycle time. For the current instruction to be evaluated, based on the execution time of the current instruction to be evaluated and the remaining available time, it is determined whether the current instruction to be evaluated meets the constraint conditions; When the instruction to be evaluated is added to the selected instruction set, the execution time of the instruction to be evaluated is subtracted from the remaining available time to obtain the updated remaining available time. The updated remaining available time is used to determine the constraint conditions for the next atomic detection instruction.

[0009] According to a device performance testing method provided by the present invention, determining whether the current instruction to be evaluated satisfies the constraint condition based on the execution time of the current instruction to be evaluated and the remaining available time includes: Determine whether the execution time of the currently evaluated instruction is not greater than the remaining available time to obtain a first determination result; Determine whether the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship to obtain a second determination result; If the first judgment result indicates that the execution time of the current instruction to be evaluated is not greater than the remaining available time, and the second judgment result indicates that the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship, then the current instruction to be evaluated is determined to satisfy the constraint condition.

[0010] According to a device performance testing method provided by the present invention, after completing the judgment of all the atomic detection instructions in the sorted candidate detection instruction sequence, the method further includes: Determine whether the remaining available time is greater than zero; If the remaining available time is greater than zero, atomic detection instructions that have not been added to the selected instruction set and whose execution time is not greater than the remaining available time are selected from the candidate detection instruction set and are used as the instruction set to be filled. According to the order of execution time from shortest to longest, each atomic detection instruction in the instruction set to be filled is taken as the current instruction to be filled, so as to determine whether the current instruction to be filled satisfies the constraint condition. If the conditions are met, the current instruction to be filled is added to the selected instruction set as the target atomicity detection instruction, and the execution time of the current instruction to be filled is deducted from the remaining available time. If the conditions are not met, the current instruction to be filled is skipped until the remaining available time is insufficient to accommodate the execution time of any atomicity detection instruction in the instruction set to be filled. Arrange the target atomization detection instructions in the selected instruction set after filling according to the order of addition, and update the target detection instruction sequence.

[0011] According to a device performance testing method provided by the present invention, after sending the target testing instruction sequence to the target device, the method further includes: Receive execution result data fed back by the target device for the target atomization detection instructions that have been executed in the target detection instruction sequence; Determine whether the execution result data meets the preset exception triggering conditions; If the execution result data meets the abnormal triggering condition, based on the execution result data, a depth detection instruction corresponding to the abnormal triggering condition is obtained from a preset depth detection instruction library; The depth detection instruction is inserted after the executed target atomization detection instruction in the target detection instruction sequence; Based on the execution time of the depth detection instruction and the execution time of the unexecuted target atomic detection instructions in the target detection instruction sequence, target atomic detection instructions that meet the preset removal rules are removed from the target detection instruction sequence to obtain an updated target detection instruction sequence. Send the depth detection instruction and the unexecuted target atomization detection instruction from the updated target detection instruction sequence to the target device to control the target device to continue performing detection according to the updated target detection instruction sequence.

[0012] According to a device performance testing method provided by the present invention, the step of obtaining a candidate testing instruction set includes: Obtain fault mode information associated with the target device; Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomized detection instructions corresponding to the fault mode information. The knowledge graph pre-stores the mapping relationship between fault modes and atomized detection instructions. All the retrieved atomized detection instructions are summarized to obtain the candidate detection instruction set.

[0013] According to a device performance testing method provided by the present invention, the knowledge graph pre-stores the mapping relationship between product model information and failure mode information and atomized testing instructions, and the failure mode information includes the target product model information and target failure mode information of the target device. Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: Use the target product model information and the target failure mode information as the joint search key; Search the knowledge graph for product model information and failure mode information that match the joint search key, and use the atomized detection instructions mapped by the found product model information and failure mode information as the atomized detection instructions corresponding to the failure mode information.

[0014] According to a device performance testing method provided by the present invention, the knowledge graph pre-stores multiple known failure mode information and atomized detection instructions corresponding to each of the known failure mode information, and the failure mode information includes the target failure mode information of the target device; Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: Feature extraction is performed on the target failure mode information to obtain the target embedding vector corresponding to the target failure mode information; Obtain the known embedding vectors corresponding to each known failure mode information in the knowledge graph; Calculate the semantic similarity between the target embedding vector and each of the known embedding vectors; The known failure mode information corresponding to the known embedding vector whose semantic similarity satisfies the preset similarity condition is used as the similar failure mode information; The atomized detection instructions mapped to the similar failure mode information in the knowledge graph are used as the atomized detection instructions corresponding to the failure mode information.

[0015] According to a device performance testing method provided by the present invention, the knowledge graph pre-stores fault phenomenon nodes, component nodes, functional module nodes, and atomized detection instruction nodes. The fault phenomenon nodes are associated with the component nodes, the component nodes are associated with the functional module nodes, and the functional module nodes are associated with the atomized detection instruction nodes. The fault mode information includes the target fault phenomenon information of the target device. Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: In the knowledge graph, identify the target fault phenomenon node that matches the target fault phenomenon information; Based on the association between the target fault phenomenon node and each of the component nodes, the target component node associated with the target fault phenomenon node is determined; Based on the association between the target component node and each of the functional module nodes, the target functional module node associated with the target component node is determined; Based on the association between the target functional module node and each of the atomization detection instruction nodes, the target atomization detection instruction node associated with the target functional module node is determined; The atomic detection instruction corresponding to the target atomic detection instruction node is used as the atomic detection instruction corresponding to the fault mode information.

[0016] According to a device performance testing method provided by the present invention, before obtaining the candidate testing instruction set, the method further includes: Obtain historical fault mode information associated with each of the atomic detection instructions in the preset instruction library; Obtain the severity parameter, occurrence frequency parameter, and time decay factor corresponding to the historical fault mode information. The severity parameter is used to characterize the severity of the fault consequences corresponding to the historical fault mode information. The occurrence frequency parameter is used to characterize the occurrence frequency of the historical fault mode information within a preset statistical period. The time decay factor is used to characterize the influence weight of the occurrence time of the historical fault mode information on the current moment. The risk weight value corresponding to the atomized detection command is obtained by weighting the severity parameter, the occurrence frequency parameter, and the time decay factor. The risk weight value is associated with and stored in relation to the atomization detection instruction.

[0017] According to a device performance testing method provided by the present invention, the step of obtaining the time decay factor corresponding to the historical fault mode information includes: Obtain the fault occurrence time corresponding to the historical fault mode information; Calculate the time difference between the current moment and the time when the fault occurred; A negative exponential calculation is performed based on a preset attenuation coefficient and the time difference to obtain the time attenuation factor corresponding to the historical fault mode information. The time attenuation factor decreases as the time difference increases.

[0018] According to a device performance testing method provided by the present invention, the step of obtaining historical fault mode information associated with each of the atomic testing instructions in a preset instruction library includes: Acquire multi-source quality feedback data, which includes at least one of return / exchange work order data, maintenance record text data, and batch quality event data; The multi-source quality feedback data is cleaned to obtain cleaned quality feedback data; The cleaned quality feedback data is input into the text processing model to obtain multiple candidate fault mode information and the confidence score corresponding to each candidate fault mode information after the text processing model performs entity recognition processing and relation extraction processing in sequence. The entity recognition process is used to identify product model entities, component entities, and fault phenomenon entities from the cleaned quality feedback data, and the relationship extraction process is used to establish the association relationship between the product model entities, the component entities, and the fault phenomenon entities. The candidate fault mode information associated with the atomization detection command among the candidate fault mode information whose confidence scores meet the preset confidence conditions is used as the historical fault mode information.

[0019] According to a device performance testing method provided by the present invention, after associating and storing the risk weight value with the atomized detection instruction, the method further includes: Determine whether the preset risk weight value update triggering conditions are met. The risk weight value update triggering conditions include timed triggering conditions and / or event triggering conditions. The timed triggering condition is that the time interval between the current time and the last time the risk weight value update was performed reaches a preset update cycle. The event triggering condition is that a new batch quality event that meets the preset event level is detected in the batch quality event data. If the timed triggering condition or the event triggering condition is met, the risk weight value corresponding to each atomic detection instruction in the candidate detection instruction set is recalculated.

[0020] According to a device performance testing method provided by the present invention, the atomized testing instruction includes one or more of the following: instruction identifier field, input parameter field, output result field, and dependency relationship field; The instruction identifier field is used to uniquely identify the atomization detection instruction, and the instruction identifier field contains a product category code and a function type code; The input parameter field is used to store the standardized input parameters required when the atomization detection instruction is executed; The output result field is used to store the standardized output result fed back after the atomization detection instruction is executed; The dependency field is used to store the execution association between the atomic detection instruction and other atomic detection instructions.

[0021] According to a device performance testing method provided by the present invention, the target device is equipped with a general instruction execution engine, which includes a communication protocol parsing module, an instruction scheduling module, a status monitoring module, and a result feedback module. Sending the target detection command sequence to the target device to control the target device to perform detection according to the target detection command sequence includes: The target detection instruction sequence is sent to the communication protocol parsing module of the target device through a preset communication protocol, so as to control the communication protocol parsing module to perform protocol parsing on the target detection instruction sequence to obtain each target atomization detection instruction; The instruction scheduling module is controlled to schedule each of the target atomization detection instructions to the corresponding functional module driver unit of the target device for execution in the order of the target detection instruction sequence. The status monitoring module is controlled to monitor the execution status of each target atomization detection instruction in real time, and triggers a preset exception handling process when an abnormal execution status is detected. The control result feedback module collects the execution result data of each of the target atomization detection instructions and sends the execution result data back through the communication protocol.

[0022] The present invention also provides a device for testing equipment performance, comprising: The cycle time acquisition module is used to obtain the upper limit of the cycle time issued by the production line where the target equipment is located; The instruction acquisition module is used to acquire a candidate detection instruction set, which contains multiple independent atomic detection instructions. Each atomic detection instruction is configured with a risk weight value, execution time, and execution correlation. The sequence generation module is used to maximize the total risk weight value as the optimization objective, and to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence, subject to the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time and that each selected target atomic detection instruction satisfies the execution association relationship. The instruction issuing module is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0023] This invention also provides an intelligent agent for device performance detection, mainly comprising: The data processing layer is used to obtain the upper limit of the cycle time issued on the production line where the target device is located; and to obtain a candidate detection instruction set, which contains multiple independent atomic detection instructions, each of which is configured with a risk weight value, execution time and execution correlation. The knowledge reasoning layer is used to maximize the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence. An execution feedback layer is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0024] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the device performance detection method as described above.

[0025] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the device performance detection method as described above.

[0026] The device performance testing method, apparatus, intelligent agent, electronic device, and storage medium provided by this invention decouple the testing instructions atomically and generate the testing sequence based on multi-constraint dynamic programming, thereby decoupling the testing logic from the electronic control firmware, supporting rapid iteration of the testing strategy, and maximizing the quality risk coverage within a limited cycle time. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating existing equipment performance testing methods.

[0029] Figure 2 This is a flowchart illustrating the equipment performance testing method provided by the present invention.

[0030] Figure 3 This is a schematic diagram of the process for generating target detection instruction sequences provided by the present invention.

[0031] Figure 4 This is a flowchart illustrating the process of obtaining a candidate detection instruction set provided by the present invention.

[0032] Figure 5This is a schematic diagram of the process for determining risk weight values ​​provided by the present invention.

[0033] Figure 6 This is a schematic diagram of the decoupling architecture for atomized detection instructions provided by the present invention.

[0034] Figure 7 This is a schematic diagram of the overall workflow of the equipment performance testing method provided by the present invention.

[0035] Figure 8 This is a schematic diagram of the device performance testing apparatus provided by the present invention.

[0036] Figure 9 This is a schematic diagram of the structure of the intelligent agent for device performance detection provided by the present invention.

[0037] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0039] It should be noted that, in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Those skilled in the art will understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0040] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more.

[0041] Figure 1This is a flowchart illustrating existing equipment performance testing methods, such as... Figure 1 As shown, the performance testing of target equipment, such as refrigerators and washing machines, before final assembly on the production line is typically conducted in the following way: Engineers manually determine a set of performance testing items from all possible testing items (e.g., testing item 1, testing item 2, testing item 3, ..., testing item N) based on the manually compiled statistics of high-risk quality items and the production line's inspection cycle time constraints. Subsequently, electrical control developers develop and test the corresponding performance testing electrical control program for the determined item combination, and then compile and burn the developed program into the equipment's electrical control firmware. Before the equipment rolls off the production line, the performance testing mode is activated via button combinations or communication protocols. The electrical control firmware then runs the various testing conditions corresponding to the item combination in a pre-defined order to complete the testing of the equipment's main functional modules. When subsequent changes occur, such as process changes, incoming material changes, or design changes, the above process can be repeated to update the testing item combination and the corresponding electrical control program.

[0042] However, long-term practice has revealed that the existing equipment performance testing methods described above have at least the following technical shortcomings: On the one hand, once the combination of test items is determined, it is hard-coded into the electronic control firmware of the equipment through the electronic control program. During the mass production process, the composition of the test items and the execution order of each test item are locked in a static state. It is impossible to dynamically optimize and select based on the actual quality risk of each test item during the production line operation. As a result, the limited production line cycle time resources are consumed uniformly on the experience-preset test items, making it difficult to accurately tilt the test resources towards the truly high-quality risk items.

[0043] On the other hand, under the hard constraint of production line cycle time, due to the significant differences in execution time of different inspection items, and the complex coupling constraints such as execution sequence dependencies and mutual exclusion relationships among the inspection items, existing solutions simply run the inspection conditions one by one in a pre-fixed order. This fails to optimize the selection and sequence arrangement of inspection items within the feasible domain of these multiple constraints, easily leading to resource misallocation where low-quality, high-risk inspection items are excessively occupied by cycle time, while high-quality, high-risk inspection items are forced to be abandoned due to insufficient remaining time. This significantly limits the overall defect detection capability within the limited cycle time. The aforementioned sequence dependencies may mean that some inspection items can only be started after the preceding conditions are completed, while mutual exclusion relationships may mean that some inspection items cannot be executed simultaneously due to shared functional modules or conflicting conditions.

[0044] In view of this, embodiments of the present invention provide a device performance testing method, apparatus, intelligent agent, electronic device, and storage medium, which can fundamentally overcome the above-mentioned defects of the prior art.

[0045] Before detailing the specific process of the equipment performance testing method provided in this embodiment, it should be noted that this method can be executed by various electronic devices with data computing and instruction scheduling capabilities. In a real industrial manufacturing environment, the executing entity can be a production line industrial control computer deployed next to the workstation, a dedicated testing host computer, a cloud server of the Manufacturing Execution System (MES), or an edge computing node integrated inside the testing equipment. For the sake of brevity and coherence, the control system will be consistently used as the executing entity of this testing method in the following detailed embodiments.

[0046] Figure 2 This is a flowchart illustrating the equipment performance testing method provided by the present invention, as shown below. Figure 2 As shown, the device performance testing method includes the following steps: Step 11: Obtain the upper limit of the cycle time of the production line where the target equipment is located.

[0047] The target equipment refers to the physical product undergoing functional verification, such as a refrigerator, washing machine, or air conditioner that has just completed its final assembly process on the production line. During the final assembly and off-line stage, there is a clear physical time limit for the dwell time of the target equipment at the current workstation. Due to the continuous operation requirements of industrial production lines, the dwell time of the target equipment at the testing station is strictly constrained; this time constraint is the upper limit of the cycle time.

[0048] In practical implementation, there are several ways to obtain the cycle time upper limit. For example, the standard cycle time configuration parameters for the current batch of products can be retrieved directly from the factory's MES via industrial Ethernet; alternatively, dynamic cycle pulse signals issued by the production line's programmable logic controller (PLC) can be received in real time via a serial interface or fieldbus; furthermore, barcodes or radio frequency identification (RFID) tags on the target equipment can be read using a barcode scanner at the workstation, and the pre-configured cycle time upper limit can be matched in a local mapping table. In certain debugging scenarios, engineers can also manually input or modify the cycle time upper limit through a human machine interface (HMI).

[0049] Step 12: Obtain a candidate detection instruction set, which contains multiple independent atomic detection instructions. Each atomic detection instruction is configured with a risk weight value, execution time, and execution correlation.

[0050] In one alternative embodiment, the candidate detection instruction set can be obtained by directly reading the full set of test instructions corresponding to the model of the target device from the local database, or by dynamically retrieving it from the cloud quality management platform via a wireless network, or by generating it in real time based on the specific status file passed from the previous process.

[0051] Specifically, the acquired candidate detection instruction set contains multiple independent atomic detection instructions. Each atomic detection instruction refers to control code that breaks away from traditional continuous hard-coded logic and is decomposed into the smallest independently controllable execution unit. Each independent atomic detection instruction typically corresponds to the drive control of a single physical actuator, such as opening a water inlet valve, starting a motor, or acquiring the status of a single sensor, such as reading the temperature sensor value. It should be noted that each atomic detection instruction generally does not contain complex business judgments or hard timing waits.

[0052] In this embodiment, each independent atomic detection instruction is configured with multi-dimensional attribute parameters, mainly including: risk weight value, execution time, and execution correlation.

[0053] The risk weight value quantifies the severity or probability of quality defects covered by each atomized detection command; a higher risk weight value indicates a more critical detection. Execution time characterizes the standard time required for the target device to fully execute a single action at the physical level. Execution dependencies are safety protection logic and topological constraints set to prevent damage to the underlying hardware. These can include execution order dependencies (e.g., the "water injection" command must be executed before the "heating" command) and execution mutual exclusion relationships (e.g., "water inlet valve open" and "water outlet valve open" have a physical conflict within the same time slice and must never be executed simultaneously).

[0054] Step 13: With the goal of maximizing the total risk weight value, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, multiple target atomic detection instructions are selected from the candidate detection instruction set and arranged in sequence to generate a target detection instruction sequence.

[0055] Considering the upper limit of the cycle time, if it is determined that it is impossible to issue and execute all instructions in the candidate detection instruction set, a trade-off must be made under multiple constraints.

[0056] Specifically, this embodiment constructs a constrained optimization mathematical model in the background. The optimization objective of this model is set to maximize the total risk weight value, that is, to expect the final selected instruction combination to intercept as many high-risk quality risks as possible. At the same time, relevant constraints are set for the optimization mathematical model: the total execution time of all selected target atomic detection instructions must not exceed the upper limit of the production line's cycle time; when the selected target atomic detection instructions are combined, they must all satisfy the aforementioned execution correlation relationship, and there must be no logical inversion or parallel allocation of mutually exclusive instructions.

[0057] By performing calculations within the feasible region defined by the above constraints, multiple target atomization detection commands with the highest detection value can be selected from the candidate detection command set and arranged in a reasonable sequence on the time axis.

[0058] In one optional implementation, the above-mentioned constraint solving and sequence arrangement can be accomplished by calling an Integer Linear Programming (ILP) solver for exact solution, or by using metaheuristic algorithms such as genetic algorithms and simulated annealing algorithms for iterative optimization, or by using greedy heuristic algorithms for rapid arrangement. Through the above solution process, the discrete target atomized detection instructions are finally assembled and spliced ​​to generate a target detection instruction sequence that is timely, secure, reliable, and can maximize benefits within a specified time.

[0059] Step 14: Send the target detection command sequence to the target device.

[0060] In this embodiment, the generated target detection instruction sequence will be sent to the physical entity. In specific implementation, the communication link for sending the target detection instruction sequence to the target device can be sent via wired serial communication such as RS485 or RS232, or via Controller Area Network (CAN) bus, or via wireless transparent transmission using Wireless Fidelity (Wi-Fi), Bluetooth, or a dedicated factory IoT protocol.

[0061] After receiving the target detection instruction sequence, the target device's internal electronic control module can sequentially parse and schedule the underlying hardware drivers, thereby controlling the target device to execute various performance detection conditions according to the target detection instruction sequence.

[0062] The equipment performance testing method provided in this invention breaks down the originally rigid and lengthy testing program into independent atomic testing instructions with attribute parameters, changing the existing technical situation where testing items must be hard-coded and cannot be selected. Simultaneously, it transforms the performance testing process into a multi-constraint optimization problem, enabling the automatic maximization of the total risk weight value under the hard constraints of the cycle time limit and the safety constraints of execution relationships. This not only decouples the testing logic from the underlying electronic control firmware but also ensures that each scheduling precisely allocates limited production line testing time resources to the testing items with the highest quality risk. Therefore, without affecting the production line's flow rate, it significantly improves the overall defect detection capability and quality risk coverage benefits per unit testing time.

[0063] Figure 3 This is a schematic diagram of the process for generating a target detection instruction sequence provided by the present invention. Figure 3 The dashed box on the left shows the planning input constraints upon which sequence planning depends. These include candidate detection instruction sets with risk weights, production line cycle time limits, instruction pre-dependencies, and instruction mutual exclusion constraints. These planning input constraints are input in parallel to... Figure 3 The heuristic multi-constraint programming algorithm is shown in the dashed box on the right. This algorithm sequentially performs tasks including sorting by risk weight value in descending order, time checks based on remaining available time, conflict checks based on execution relationships, and filling in remaining time fragments. The final output is a sequence of target detection instructions and their execution time schedule. Figure 3 The dynamic adaptive adjustment section shown in the dashed box below is used to dynamically update the planning input constraints based on the execution result feedback after the target detection instruction sequence is issued and executed. For details of its implementation process, please refer to the corresponding embodiments below.

[0064] Combination Figure 3 As shown, the step of selecting multiple target atomized detection instructions from the candidate detection instruction set and arranging them in sequence to generate a target detection instruction sequence may specifically include the following steps: First, the atomic detection instructions in the candidate detection instruction set are sorted in descending order of risk weight values ​​to obtain a sorted sequence of candidate detection instructions.

[0065] When faced with a large and complex set of candidate detection instructions, this embodiment prioritizes all independent atomic detection instructions to ensure that the most valuable detection items are selected first. In practice, efficient sorting algorithms such as quicksort, mergesort, or heapsort can be used. The atomic detection instructions with the highest risk weight value (e.g., the quantified Risk Priority Number, RPN) are placed at the beginning of the queue, while those with lower risk weight values ​​are placed at the end. This creates a queue of candidate detection instructions ordered in descending order of quality risk severity.

[0066] Furthermore, according to the sorted sequence of candidate detection instructions, each of the atomic detection instructions is taken as the current instruction to be evaluated, and it is determined whether the current instruction to be evaluated satisfies the constraint conditions.

[0067] After sorting, this embodiment extracts each atomization detection instruction sequentially in descending order for feasibility evaluation. The instruction currently to be evaluated refers to the atomization detection instruction that is being read by the system and is in the evaluation state. Determining whether the instruction currently to be evaluated meets the constraints means verifying whether selecting the instruction would break two constraint boundaries in the optimization model.

[0068] In one alternative implementation, the constraint verification can be performed by using a pre-built conflict detection tree, or by summing the total time of the currently selected instructions and adding the time of the instruction to be evaluated to compare with the upper limit of the cycle time, so as to comprehensively determine whether the instruction to be evaluated is legal and feasible in terms of timing and physical logic.

[0069] In actual testing, if the constraints are met, the current instruction to be evaluated is added to the selected instruction set as the target atomization detection instruction; otherwise, the current instruction to be evaluated is skipped.

[0070] The selected instruction set in memory acts as a dynamically expandable safe container. If, after verification, it is determined that the instruction to be evaluated fully meets the constraints such as time consumption boundaries and execution correlation, then the instruction to be evaluated is confirmed to be safe for issuance and execution. It is then marked as a target atomicity detection instruction and loaded into the selected instruction set.

[0071] Conversely, if the current instruction to be evaluated fails the above constraint verification, for example, the current instruction to be evaluated, "maximum speed test for dehydration", requires 3 minutes, and the total time boundary reserved by the production line is insufficient to support it, then the current instruction to be evaluated will be abandoned and skipped, and the system pointer will immediately slide forward to read the next atomized detection instruction in the sorted candidate detection instruction sequence for the next round of evaluation.

[0072] Furthermore, after determining all the atomic detection instructions in the sorted candidate detection instruction sequence, the target atomic detection instructions in the selected instruction set are arranged in the order of addition to generate the target detection instruction sequence.

[0073] It should be noted that the above traversal and filtering process will continue until the last atomic detection instruction in the sorted candidate detection instruction sequence has been evaluated. At this point, all target atomic detection instructions collected in the selected instruction set not only satisfy the correlation rules for production line safety, but also fall within the constraints in terms of overall time consumption. Since these target atomic detection instructions are selected and added sequentially according to their risk weight values ​​from high to low, their addition order in the selected instruction set reflects the decreasing relationship of their importance.

[0074] Finally, the target atomization detection instructions in the selected instruction set can be solidified and spliced ​​and arranged on the timeline according to their order of addition, thereby generating the final target detection instruction sequence used to control physical devices.

[0075] This invention employs a strategy of sorting by risk weight value in descending order and performing constraint verification sequentially. This cleverly reduces the complexity of the original multi-constraint combination optimization problem into an efficient traversal selection process with linear time complexity. This avoids the risks of computational resource exhaustion and scheduling timeouts that are easily caused by conventional exhaustive algorithms when faced with massive numbers of detection items. It ensures that within the limited production line cycle time, the most critical atomic detection instructions can be sequentially allocated according to risk level. Thus, under extremely high industrial production cycles, it can quickly and stably output a detection execution sequence that approximates the global optimal risk coverage.

[0076] Based on the above embodiments, in order to accurately manage time constraints during dynamic programming, this embodiment of the invention further provides a feasible implementation method for determining whether the currently evaluated instruction satisfies the constraint conditions, which may specifically include the following steps: First, initialize the remaining available time to the maximum cycle time. Then, for the current instruction to be evaluated, based on the execution time of the current instruction to be evaluated and the remaining available time, it is determined whether the current instruction to be evaluated meets the constraint conditions; Finally, when the instruction to be evaluated is added to the selected instruction set, the execution time of the instruction to be evaluated is subtracted from the remaining available time to obtain the updated remaining available time. The updated remaining available time is used to determine the constraint conditions for the next atomic detection instruction.

[0077] At the start of the equipment performance testing process, this embodiment pre-establishes a dynamically tracking time scale. Specifically, a variable representing the remaining available time can be declared in memory beforehand, and its initial value is assigned to the aforementioned maximum cycle time. For example, if the maximum cycle time from the production line to the current workstation is 180 seconds, then in the initial stage of the algorithm, the value of the remaining available time is set to 180 seconds. This remaining available time is equivalent to a dynamically decreasing time pool, used to reflect in real time how much testing time the current target equipment can continue to accommodate at the workstation.

[0078] When reviewing each instruction to be evaluated, its configured execution time is a key factor in determining whether it can be accepted. The control logic extracts the estimated execution time configured for the instruction to be evaluated and compares and verifies it with the current remaining available time.

[0079] By comprehensively considering the proportion of the execution time relative to the current time pool and other possible related constraints, we can assess whether the current time window is sufficient to support the target device in fully executing the detection action represented by the current instruction to be evaluated, and then determine whether the current instruction to be evaluated meets the overall constraints.

[0080] Furthermore, when the currently evaluated instruction successfully passes verification and is added to the selected instruction set, it means that this atomic detection instruction will inevitably consume a portion of the target device's physical dwell time in the future. Therefore, it is necessary to simultaneously subtract the execution time of the newly added atomic detection instruction from the current remaining available time. Assuming the remaining available time is 120 seconds, and the newly added atomic detection instruction, such as a "water inlet valve water pressure test" instruction, is configured to have an execution time of 15 seconds, then after performing the subtraction operation, the updated remaining available time is 105 seconds. Subsequently, when evaluating the next atomic detection instruction in the sorted candidate detection instruction sequence, the system will use this updated 105 seconds as the new time boundary for the next round of constraint judgment.

[0081] This invention introduces a dynamically following remaining available time variable, transforming the cycle time upper limit constraint into a micro-time pool comparison during each traversal evaluation. This allows for precise control of the time balance at each production line station during sequence arrangement, providing an accurate time scale for the admission of each subsequent atomized detection instruction. This not only effectively prevents the total time consumption of the final generated sequence from exceeding the limit, but also improves the reliability and computational efficiency of constraint solving.

[0082] Based on the above embodiments, as an optional embodiment, determining whether the current instruction to be evaluated satisfies the constraint condition based on the execution time of the current instruction to be evaluated and the remaining available time includes: Determine whether the execution time of the currently evaluated instruction is not greater than the remaining available time to obtain a first determination result; Determine whether the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship to obtain a second determination result; If the first judgment result indicates that the execution time of the current instruction to be evaluated is not greater than the remaining available time, and the second judgment result indicates that the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship, then the current instruction to be evaluated is determined to satisfy the constraint condition.

[0083] When conducting a feasibility assessment, a time-based boundary check can be performed first. In this embodiment, the pre-configured execution time parameter of the instruction to be evaluated is read and compared with the remaining available time recorded by the system. The resulting Boolean value or status indicator constitutes the first judgment result. For example, if the current remaining available time is 30 seconds, and the execution time required for the instruction to be evaluated, such as a "drainage pump operation test" instruction, is 15 seconds, then it is determined that the execution time is not greater than the remaining available time, and the resulting first judgment result indicates that the instruction to be evaluated is allowed to be added within the time window. Conversely, if the execution time exceeds 30 seconds, the first judgment result indicates that the time limit has been exceeded.

[0084] In addition to the time dimension verification, this embodiment will also verify the execution logic and security topology at the physical layer. This mainly involves comparing the correlation between the current instruction to be evaluated and all target atomization detection instructions that have been successfully added to the selected instruction set. The verification conclusion obtained after the comparison is the second judgment result.

[0085] In practical implementation, this determination of correlation can encompass both prerequisite dependency checks and mutual exclusion / conflict checks. Assuming the instruction to be evaluated is a heating tube heating test, it is necessary to check whether the selected instruction set already includes prerequisite target atomization detection instructions such as a water inlet test to provide a safe water level. Simultaneously, it is also necessary to check whether the selected instruction set includes other target atomization detection instructions that are physically mutually exclusive with the heating tube heating test or may cause overload, to ensure that the selected instruction combination does not interfere with actual physical execution.

[0086] For any instruction currently under evaluation, its final confirmation requires passing both temporal and logical checks. Only when the first judgment in the temporal dimension indicates that the remaining time at the production line station is sufficient to accommodate the action corresponding to the instruction, and the second judgment in the logical dimension indicates that adding the action will not violate the execution timing rules and equipment safety limits of the selected instruction set, will the instruction currently under evaluation be ultimately determined to fully meet the aforementioned constraints. This provides a basis for decision-making for its subsequent formal addition to the selected instruction set as a target atomization detection instruction.

[0087] The embodiments of the present invention decouple the constraints into independent time judgments and execution logic judgments through the above-mentioned dual-dimensional verification mechanism. This not only makes the constraint verification process more organized and standardized, but also ensures that the final generated combination of detection instructions will not break through the strict physical time constraints of industrial production lines, and can fully comply with the safety rules of the underlying hardware operation of home appliances and other equipment. This effectively avoids the risk of equipment and component damage caused by conflicting operating conditions and improves the feasibility and execution security in complex industrial environments.

[0088] Based on the above embodiments, after the aforementioned main loop of risk weight-based descending evaluation, due to the differences in execution time of each testing instruction, some time fragments that cannot accommodate large, high-risk test items often remain. To make the best use of the production line cycle time limit, further consideration is needed... Figure 3 As shown, after completing the judgment of all the atomic detection instructions in the sorted candidate detection instruction sequence, the embodiment of the present invention further performs the remaining time fragment filling operation, which mainly includes: Determine whether the remaining available time is greater than zero; If the remaining available time is greater than zero, atomic detection instructions that have not been added to the selected instruction set and whose execution time is not greater than the remaining available time are selected from the candidate detection instruction set and are used as the instruction set to be filled. According to the order of execution time from shortest to longest, each atomic detection instruction in the instruction set to be filled is taken as the current instruction to be filled, so as to determine whether the current instruction to be filled satisfies the constraint condition. If the conditions are met, the current instruction to be filled is added to the selected instruction set as the target atomicity detection instruction, and the execution time of the current instruction to be filled is deducted from the remaining available time. If the conditions are not met, the current instruction to be filled is skipped until the remaining available time is insufficient to accommodate the execution time of any atomicity detection instruction in the instruction set to be filled. Arrange the target atomization detection instructions in the selected instruction set after filling according to the order of addition, and update the target detection instruction sequence.

[0089] After the first round of main loop traversal and filtering, this embodiment first checks the baseline status of the time pool. Specifically, it reads the current remaining available time value and determines whether it is still a positive number greater than zero. If the remaining available time is equal to zero or close to zero, it means that the production line cycle limit has been fully exhausted and no further replenishment is needed; if the remaining available time is greater than zero, it indicates that the target equipment still has an idle dwell period at the testing station, and this time can be used to perform some minor testing actions.

[0090] Specifically, when it is confirmed that there are still time fragments available for use, this embodiment will rescan the original candidate detection instruction set in the background. At this time, the screening criteria become: looking for atomic detection instructions that were rejected in the first round of traversal due to their relatively low risk weight and insufficient remaining time, and whose execution time can fit into the current very short remaining available time. The atomic detection instructions that meet the above conditions are extracted and aggregated to form a set of instructions to be filled.

[0091] To cram as many detection actions as possible into the extremely limited fragmented time, this embodiment appropriately transforms the sorting strategy. Specifically, the instructions within the instruction set to be filled can be arranged in ascending order according to their execution time, so that instructions with extremely short execution times are placed at the front. Subsequently, each atomic detection instruction is extracted sequentially along this ascending queue as the current instruction to be filled, and the aforementioned constraint verification logic is reused to focus on checking whether there are physical conflicts or missing dependencies between the current instruction to be filled and existing instructions in the selected instruction set, in order to determine whether it is safe and feasible.

[0092] If the instruction to be filled passes the logical check of the association constraint, it is officially assigned the identity of the target atomicity detection instruction and appended to the end of the selected instruction set, while its execution time continues to be subtracted from the remaining available time. If the target atomicity detection instruction logically conflicts with the selected instructions, it is skipped, and the next instruction with slightly longer execution time in the instruction set to be filled is examined. The above cycle of attempting to fill and deducting time will continue until the time pool is completely exhausted, that is, the updated remaining available time is so short that even the shortest instruction in the instruction set to be filled cannot be accommodated.

[0093] After the aforementioned time fragment filling cycle is fully completed, the selected instruction set not only includes the high-risk main test items selected in the first round, but also several extremely short auxiliary test items added in the second round. Finally, this filled and expanded selected instruction set is solidified, and time-series splicing is performed according to the final order in which all instructions were added, thereby updating and upgrading the original target detection instruction sequence.

[0094] The equipment performance testing method provided in this invention, under the hard constraint of fixed cycle time, cleverly introduces a remaining time fragment filling mechanism after arranging high-risk items in the main cycle. The short-duration rejection instructions are evaluated and filled in ascending order of duration, so that the equipment performs as many testing actions as possible within a limited dwell period. This not only ensures that high-risk quality defects are intercepted first at the main level, but also further expands the coverage of overall quality testing at the secondary level, effectively avoiding the hidden waste of cycle time on industrial production lines.

[0095] Based on the above embodiments, as an optional embodiment, after sending the target detection command sequence to the target device, a dynamic adaptive adjustment mechanism based on real-time detection feedback is further provided (i.e., Figure 3 The dynamically adaptive adjustment real-time detection and feedback mechanism shown may specifically include the following steps: First, the execution result data fed back by the target device for the target atomization detection instructions that have been executed in the target detection instruction sequence is received.

[0096] As the target device executes various tests step by step according to the predetermined sequence of target detection instructions, its internal microcontroller or sensor network collects the physical operating status of each functional module in real time. For example, after executing a "dehydration motor drive test" instruction, the target device can upload parameters such as motor phase current, speed ramp-up curve, or vibration frequency collected in real time. These parameters can objectively reflect the physical parameters or status codes that reflect the quality of the specific action and the health of the underlying devices, thus constituting the aforementioned execution result data.

[0097] Further, it is determined whether the execution result data meets the preset exception triggering conditions.

[0098] Specifically, after obtaining real-time execution result data, online verification of fault signs can be performed in the algorithm backend. The preset anomaly trigger conditions can be predefined standard numerical ranges (such as current upper limit thresholds) or standard signal fluctuation characteristics. By comparing the received execution result data with the preset anomaly trigger conditions, it is possible to identify whether the target device exhibits abnormal signs such as waveform distortion, response delay, or excessive power consumption when performing a certain action. If the comparison passes, it indicates that the target device is operating smoothly; if a deviation occurs, it indicates a potential early failure in a specific direction.

[0099] Furthermore, if the execution result data meets the abnormal triggering condition, a depth detection instruction corresponding to the abnormal triggering condition is obtained from a preset depth detection instruction library based on the execution result data.

[0100] When a certain indicator is detected to meet the aforementioned abnormal triggering conditions, this embodiment will not directly determine that the target device is unqualified. Instead, it will trigger a secondary verification mechanism based on this judgment result. For example, this embodiment will introduce an independently constructed preset deep detection instruction library, which stores high-frequency, high-intensity dedicated detection instructions specifically used for locating various difficult faults and extreme verifications. The control logic will directly map and retrieve the instruction from this deep detection instruction library according to the specific abnormality type that is currently triggered, thereby accurately matching the dedicated verification instruction for the suspected fault as the deep detection instruction. For example, for the aforementioned current abnormality, the "motor stall limit test" instruction will be retrieved.

[0101] Furthermore, the depth detection instruction is inserted after the executed target atomization detection instruction in the target detection instruction sequence.

[0102] To achieve rapid verification, this embodiment will change the original pre-arranged remaining detection plan. For example, in the memory state machine, the newly acquired deep detection instruction will be attached after the time node of the target atomization detection instruction that is currently abnormal, so as to take priority to preempt the execution right of the next time slice, thereby realizing dynamic insertion on the detection sequence execution chain.

[0103] Furthermore, based on the execution time of the depth detection instruction and the execution time of the unexecuted target atomic detection instructions in the target detection instruction sequence, target atomic detection instructions that meet the preset removal rules are removed from the target detection instruction sequence to obtain an updated target detection instruction sequence.

[0104] Since the introduction of unplanned deep inspection tasks will inevitably consume part of the original production line cycle time budget, this embodiment will release an equivalent amount of time from the unexecuted planned tasks to prevent workstation delays. At this time, the execution time required for the newly introduced deep inspection instructions will be evaluated, and those target atomic inspection instructions that are at the end of the sequence and have not yet been executed will be traversed. Based on preset elimination rules, for example, instructions with the lowest risk weight and the longest execution time among the remaining instructions with low correlation will be prioritized. The selected instructions to be eliminated will be removed from the original target inspection instruction sequence until the total time freed up can compensate for the time consumed by the newly added deep inspection tasks, and then the rebalanced updated target inspection instruction sequence will be output.

[0105] Finally, the depth detection instruction and the unexecuted target atomization detection instruction from the updated target detection instruction sequence are sent to the target device to control the target device to continue performing detection according to the updated target detection instruction sequence.

[0106] Once the new target detection instruction sequence has completed dynamic reconstruction and time balancing, the control logic will continue to transmit this revised instruction list downstream. At this point, the target device's response logic will shift, prioritizing the allocation of underlying hardware resources to execute the newly added deep detection scenario, and then orderly resuming the execution of the remaining routine test items after successful verification.

[0107] The equipment performance testing method provided in this embodiment is no longer a one-way channel that only issues code once, but forms a closed-loop detection circuit. It captures latent quality fluctuations in equipment in real time by pre-setting abnormal trigger conditions, and injects targeted deep testing instructions when necessary. Simultaneously, it ensures that the cycle time limit is met by eliminating low-risk preset items. This allows for the instantaneous reallocation of testing computing power and time resources to address sudden and difficult risks on-site while maintaining the normal cycle time of the production line. This significantly improves the accuracy and first-pass detection capability of complex equipment such as home appliances in identifying hidden manufacturing defects.

[0108] Figure 4 This is a flowchart illustrating the process of obtaining a candidate detection instruction set provided by the present invention. The dashed box on the left shows the retrieval input, namely the fault mode information associated with the target device. The dashed box in the middle shows the FMEA knowledge graph reasoning engine, which displays three retrieval methods that can be selected or combined: Method 1, exact match retrieval; Method 2, semantic similarity reasoning; and Method 3, transitive reasoning. The atomic detection instructions retrieved by the three retrieval methods are summarized and deduplicated to form the candidate detection instruction set shown on the right.

[0109] Combination Figure 4 As shown, to achieve automatic and accurate mapping from quality issues to underlying detection instructions, this embodiment provides an optional implementation method for obtaining candidate detection instruction sets, mainly including the following steps: First, obtain the fault mode information associated with the target device.

[0110] Before determining the performance testing tone of the target device, this embodiment will first identify the quality shortcomings and potential failure manifestations that the target device may have at the physical level. The obtained failure mode information is used to characterize the abnormal phenomena and characteristics that may occur or have occurred in the past of the target device, such as damage to specific components, failure of functional modules or system error codes.

[0111] In practice, the product serial number (SN) of the target equipment can be read by the workstation barcode scanner to parse the product model and production batch. Then, the product model can be used to query and extract the potential failure mode and effect analysis (FMEA) records or historical defect list that are highly correlated with it from the factory's quality information management system. These records and lists are used as failure mode information corresponding to the target equipment.

[0112] Furthermore, based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomized detection instructions corresponding to the fault mode information. The knowledge graph pre-stores the mapping relationship between fault modes and atomized detection instructions.

[0113] A knowledge graph (KG) is a network topology database formed by structured modeling of historical quality data, product design documents, and test engineering experience. This knowledge graph predefines various entity nodes, such as product model nodes, functional module nodes, component nodes, failure mode nodes, and atomic test instruction nodes, as well as directed edges between these entity nodes. These directed edges constitute the aforementioned mapping relationships, used to represent network topology logic such as "a failure mode is caused by a certain component," "a functional module contains a certain component," and "a certain atomic test instruction is used to verify a certain failure mode."

[0114] To transform abstract fault manifestations into executable low-level actions, this embodiment introduces a pre-constructed knowledge graph as the basis for inference. After acquiring fault mode information, it is input into the knowledge graph's inference engine as a retrieval input condition. Starting from the node corresponding to the fault mode information, the inference engine performs graph traversal and path retrieval along pre-stored mapping relationships within the knowledge graph network. By searching layer by layer along the associated edges, it can ultimately extract atomized detection instruction nodes that can verify, reproduce, or detect the fault mode information, thereby obtaining multiple atomized detection instructions that have a strict correspondence with the fault mode information in engineering logic.

[0115] Finally, all the retrieved atomized detection instructions are summarized to obtain the candidate detection instruction set.

[0116] As multiple search paths in the knowledge graph are traversed, a large number of scattered atomic detection instructions are obtained. Considering that the same atomic detection instruction may detect multiple different fault modes simultaneously, duplicate instructions may be output under different search paths. Therefore, this embodiment will aggregate all the retrieved atomic detection instructions, remove duplicate instruction records, and load all deduplicated instruction instances into the same data set, thereby constructing a complete and unique candidate detection instruction set to provide a basic material library for subsequent sequence generation and arrangement.

[0117] This invention transforms the traditional manual specification process of detection items based on human experience into an automatic mapping and graph reasoning process based on structured knowledge graphs. It can analyze the causal relationship and verification logic between fault phenomena and underlying software and hardware driver instructions, ensuring that the generated candidate detection instruction set can cover all known and potential quality risks currently faced by the target device without omission. This fundamentally reduces the dependence of the detection scheme on the engineer's personal experience and improves the consistency and scientific nature of quality control rules.

[0118] Continue to combine Figure 4 As shown (i.e.) Figure 4 Method 1: Precise Matching Retrieval. Based on the above embodiments, the knowledge graph pre-stores the mapping relationship between product model information and failure mode information and atomized detection instructions, respectively. The failure mode information includes the target product model information and target failure mode information of the target device.

[0119] Based on this, the present invention further provides a specific implementation method for obtaining multiple atomized detection instructions corresponding to the fault mode information by performing mapping retrieval in a pre-constructed knowledge graph based on the fault mode information, which mainly includes the following implementation steps: Use the target product model information and the target failure mode information as the joint search key; Search the knowledge graph for product model information and failure mode information that match the joint search key, and use the atomized detection instructions mapped by the found product model information and failure mode information as the atomized detection instructions corresponding to the failure mode information.

[0120] For the pre-constructed knowledge graph mentioned above, this embodiment pre-stores a rich set of graph nodes and connecting edges when performing FMEA knowledge graph modeling. Specifically, the entity types in this knowledge graph include at least product models, functional modules, components, failure modes, detection conditions, and atomic commands; the relationship types can cover inclusion relationships, causal relationships, correspondence relationships, verification relationships, and causal relationships; each entity node in the graph can also be configured with attributes such as severity, frequency of occurrence, detectability, risk priority value, and detection duration. Based on the above multi-dimensional graph modeling, the resulting knowledge graph pre-stores the mapping relationships between product model information and failure mode information and atomic detection commands, respectively. Correspondingly, the aforementioned acquired failure mode information specifically includes the target product model information and target failure mode information of the target device.

[0121] When performing multi-dimensional retrieval reasoning, the target product model information and the target failure mode information carried in the failure mode information are extracted, and the target product model information and the target failure mode information are associated and concatenated as a joint retrieval key for precise targeting in the knowledge graph network.

[0122] Furthermore, by utilizing the generated joint search key, the graph query language can be driven to perform precise node comparison and search in the knowledge graph, locating the product model information node and failure mode information node that perfectly match the joint search key.

[0123] Subsequently, based on the topological edges such as the correspondence or verification relationship defined above, the atomic detection instruction nodes associated with the found product model information and failure mode information are directly mapped and extracted along the graph path.

[0124] Finally, the atomization detection instructions mapped to the above nodes are directly used as the atomization detection instructions corresponding to the fault mode information.

[0125] By using the precise matching and retrieval method described above, which directly maps detection instructions to product models and failure modes, this embodiment establishes a hard association between specific failure characteristics and underlying execution actions. This enables the direct identification of the most accurate verification action within a very short query cycle for known and deterministic quality defects on specific product models. This effectively avoids the computational power consumption caused by traversing a massive instruction library, ensuring targeted positioning and 100% detection instruction coverage for high-frequency deterministic failure modes that are already clearly defined in engineering experience.

[0126] Combination Figure 4 As shown (i.e.) Figure 4Method 2: Semantic Similarity Reasoning. Based on the above embodiments, as an optional embodiment, the knowledge graph pre-stores multiple known failure mode information and atomic detection instructions corresponding to each of the known failure mode information. The failure mode information includes the target failure mode information of the target device.

[0127] Building upon this, this embodiment further provides an implementation method that, based on the fault mode information, performs mapping retrieval in a pre-constructed knowledge graph to obtain multiple atomized detection instructions corresponding to the fault mode information, mainly including but not limited to: Feature extraction is performed on the target failure mode information to obtain the target embedding vector corresponding to the target failure mode information; Obtain the known embedding vectors corresponding to each known failure mode information in the knowledge graph; Calculate the semantic similarity between the target embedding vector and each of the known embedding vectors; The known failure mode information corresponding to the known embedding vector whose semantic similarity satisfies the preset similarity condition is used as the similar failure mode information; The atomized detection instructions mapped to the similar failure mode information in the knowledge graph are used as the atomized detection instructions corresponding to the failure mode information.

[0128] In practical industrial applications, fault descriptions are often presented as unstructured natural language text. In order to achieve computer-readable mathematical comparison, this embodiment can call a pre-trained language model, such as Bidirectional Encoder Representations from Transformers (BERT) or word vector transformation algorithm, to perform deep semantic feature extraction on the target failure mode information expressed in text, and implicitly map it into a high-dimensional continuous numerical vector space representation, that is, generate the target embedding vector.

[0129] The known failure mode information accumulated in the knowledge graph has undergone the same text vectorization process during the node entry or graph construction stage, and the generated known embedding vectors are stored in the graph database as the basic attributes of the nodes. At this point, it can be directly extracted and loaded into the computing memory.

[0130] After obtaining the high-dimensional vector representation between the target embedding vector and each of the known embedding vectors, spatial geometric operations are performed in the vector space, such as calculating the cosine similarity, Euclidean distance, or Manhattan distance between the target embedding vector and each of the known embedding vectors, to quantitatively evaluate the degree of contextual closeness between the emerging failure modes currently faced by the target device and the known failure modes accumulated in the historical map in terms of physical failure representation and potential mechanism.

[0131] The preset similarity condition can be specifically a pre-defined lower limit of empirical similarity values, such as a cosine similarity value greater than 0.85, or set as the nearest neighbor ranking requirement in vector distance sorting. In this embodiment, known embedding vectors that meet the preset similarity condition will be extracted. Their semantic similarity indicates that the historical failure mechanisms they record are highly similar to the current situation. Therefore, the historical records pointed to by these selected known embedding vectors are identified as similar failure mode information.

[0132] Finally, by querying the network topology of the knowledge graph, the atomic detection instruction nodes mapped outward from the similar failure mode information nodes can be extracted, and the extracted atomic detection instructions can be used as atomic detection instructions corresponding to the failure mode information, thereby completing the reasoning transformation from abstract failure to specific action.

[0133] The equipment performance testing method provided in this invention, through the semantic similarity calculation and reasoning based on embedded vectors, breaks through the limitations of traditional database technology based on strict keyword comparison for retrieval. This enables the automatic use of natural language processing underlying logic to mine historical experience solutions that are highly similar in physical failure mechanism when facing new product models in the early stages of mass production, or when new failure scenarios appear on the production line with non-standardized maintenance terminology and no identical fault descriptions have been recorded. This allows for the generalization and deduction of reasonable associated detection actions, effectively compensating for matching omissions caused by differences in textual expression, and greatly enhancing the generalization detection capability for unknown or sudden defects.

[0134] Combination Figure 4 As shown, regarding the process of obtaining the candidate detection instruction set provided in the above embodiments, in addition to the aforementioned precise matching retrieval and semantic vector-based similarity reasoning implementation, this embodiment of the invention also provides another implementation method that utilizes the deep network topology of knowledge graphs for transitive reasoning (i.e. Figure 4 Method 3: Transitive reasoning.

[0135] In this application scenario, the knowledge graph pre-stores entity nodes of different granularities, specifically covering fault phenomenon nodes describing equipment anomalies, component nodes describing specific physical hardware, functional module nodes describing the equipment's macroscopic subsystems, and the lowest-level atomized detection instruction nodes. Furthermore, these fault phenomenon nodes are strictly linked at each level based on the hardware architecture and engineering causal logic of the physical product. For example, a "cause / characterization" relationship is established between fault phenomenon nodes and component nodes; a composition / membership relationship is established between component nodes and functional module nodes; and a "verification / driving" relationship is established between functional module nodes and atomized detection instruction nodes. Correspondingly, the aforementioned acquired fault mode information specifically includes the most intuitive target fault phenomenon information recorded by the target equipment at the production line or market end.

[0136] During transitive reasoning, firstly, target fault phenomenon nodes matching the target fault phenomenon information are identified in the knowledge graph. Specifically, the target fault phenomenon information is extracted, such as the recorded text "a sharp abnormal noise occurs during the washing process," and text mapping and localization are performed in the surface phenomenon network of the knowledge graph to locate the corresponding target fault phenomenon node.

[0137] Subsequently, based on the association between the target fault phenomenon node and each of the component nodes, the target component node associated with the target fault phenomenon node is determined. After locating the surface phenomenon, the graph inference engine can traverse along preset directed edges to the physical entity layer. According to the causal mapping rules accumulated in the knowledge graph, the specific physical device that may cause the above-mentioned abnormal noise is deduced, and then the graph query pointer is moved to the corresponding target component node, such as "motor bearing" or "pulsator reduction clutch".

[0138] Furthermore, based on the association between the target component node and each of the functional module nodes, the target functional module node associated with the target component node can be determined. After identifying the suspected target component, the tracing continues along the hierarchical structure of the knowledge graph. Since individual components often work collaboratively within a larger system loop, by parsing the subordinate association edges between nodes, the higher-order target functional module node where the suspected target component resides can be determined, such as the "power drive and transmission module".

[0139] Then, based on the association between the target functional module node and each of the atomic detection instruction nodes, the target atomic detection instruction node associated with the target functional module node can be determined. After reaching the module level, the graph inference engine can search for the underlying electronic control code that can activate and verify the working state of the module. Following the verification association path of the graph, the target atomic detection instruction node specifically used to test the various limit parameters of the aforementioned power drive module is retrieved, such as the "multi-stage speed climb verification instruction".

[0140] Finally, the atomic detection instruction corresponding to the target atomic detection instruction node is used as the atomic detection instruction corresponding to the fault mode information, thereby completing the entire retrieval link.

[0141] The embodiments of this invention fully leverage the structural advantages of the multi-level and network-connected structure of knowledge graphs. In fuzzy scenarios where only surface-level fault phenomena are known and there is a lack of clear failure components or mechanism determination, the invention relies on the device physical architecture and hierarchical logic accumulated in the knowledge graph to progressively trace the cause and effect, and finally uncover the underlying verification actions hidden behind the appearance. This greatly improves the thoroughness of the detection instruction retrieval scheme when facing complex system-level faults.

[0142] Figure 5 This is a schematic diagram of the process for determining risk weight values ​​provided by the present invention, such as... Figure 5 As shown, before obtaining the candidate detection instruction set, this embodiment also provides a risk quantification assessment method based on multidimensional data, which may specifically include the following steps: First, obtain historical fault mode information associated with each of the atomic detection instructions in the preset instruction library.

[0143] Before scheduling specific production, this embodiment assigns a quantitative indicator that reflects the actual interception value to each underlying control action in the candidate detection instruction set. Specifically, the system will traverse the candidate detection instruction set in the background in advance. For each independent atomic detection instruction, through the mapping relationship of the aforementioned knowledge graph or the association table of the relational database, it will trace back and extract all failure records that the atomic detection instruction has intercepted or been designed to verify in previous production or market stages. These records constitute the historical failure mode information closely bound to the instruction.

[0144] Furthermore, the severity parameter S, occurrence frequency parameter K, and time decay factor corresponding to the historical fault mode information are obtained. .

[0145] In order to evaluate the severity of a quality problem from multiple dimensions in a comprehensive and objective manner, this embodiment will retrieve three key quantitative dimension parameters.

[0146] The severity parameter S is used to characterize the severity of the consequences of the historical fault mode information. In specific implementations, the severity can be scored and assigned according to a preset fault classification standard. For example, the severity parameter can be divided into 1 to 4 level ranges, corresponding to minor faults, general faults, severe faults, and fatal or safe faults, respectively.

[0147] The frequency parameter K is used to characterize the frequency of occurrence of the historical fault mode information within a preset statistical period. For example, by extracting maintenance report data from the factory, the ratio of the actual number of occurrences of a specific fault mode within a specified statistical period (such as the last three months) to the total output during the same period can be calculated to obtain an objective frequency probability value.

[0148] Time decay factor This is used to characterize the impact weight of the occurrence time of the historical failure mode information on the current moment. Considering that quality problems often have time-sensitive and batch clustering characteristics, recent failures are often more valuable for the current production line to prevent than occasional failures that occurred a year ago. Therefore, this time decay factor is introduced to give recent failures a higher calculation weight.

[0149] Furthermore, based on the severity parameter S, the occurrence frequency parameter K, and the time decay factor... A weighted calculation is performed to obtain the risk weight value corresponding to the atomized detection command.

[0150] After obtaining the aforementioned multidimensional parameters, comprehensive mathematical operations are performed in the background data processing engine. In practical implementation, a risk priority RPN quantitative assessment model can be constructed. For example, the combined effect of the three parameters can be calculated using a product formula, namely, the extracted severity parameter S, occurrence frequency parameter K, and time decay factor. Perform a series of multiplication operations, and use the final comprehensive value as the risk weight value to quantify the interception value of the atomic detection command.

[0151] Finally, the risk weight value is associated with and stored in relation to the atomization detection instruction.

[0152] After completing the above calculations, this embodiment will use the obtained risk weight value as a dynamic attribute label and update it in the database or knowledge graph, forming a one-to-one or many-to-one binding association with the corresponding atomized detection instructions.

[0153] This embodiment creatively introduces a multi-dimensional weighted calculation of severity, objective frequency of occurrence, and time decay characteristics. This allows the calculated risk weight value to not only accurately reflect the absolute quality hazards faced by each test item, but also to keenly capture recent sudden high-frequency quality fluctuations. This ensures that newly emerging high-risk, high-frequency failure problems can be quickly assigned higher weights and prioritized in subsequent planning. This achieves dynamic tilting of testing resources towards high-risk items and effectively avoids the misallocation of limited testing cycle resources.

[0154] Combination Figure 5 As shown, based on the quality risk quantification assessment model constructed in the above embodiments, in order to further accurately quantify the timeliness impact of historical failures and specify the weighted calculation of the aforementioned multidimensional parameters, this embodiment of the invention further provides an optional implementation method for obtaining the time decay factor corresponding to the historical failure mode information, specifically including the following steps: First, obtain the fault occurrence time corresponding to the historical fault mode information.

[0155] When processing historical fault data, the first step is to pinpoint the exact time when the quality issue occurred. For example, the creation timestamp or the first repair date associated with the historical fault pattern information can be extracted from the return / exchange work order records in the Customer Relationship Management (CRM) system or from data reports on the after-sales service platform, and used as the fault occurrence time to quantify the timeliness of the issue.

[0156] Furthermore, the time difference between the current moment and the time when the fault occurred is calculated.

[0157] After obtaining the time of the fault occurrence, the current time can be read from the built-in clock or the server's Network Time Protocol (NTP). Then, through mathematical subtraction, the time span between this current time and the aforementioned fault occurrence time can be calculated, resulting in a dynamically changing time difference, denoted as . t It objectively represents the time interval between the historical failure mode and the current production line scheduling planning time.

[0158] Finally, a negative exponential operation is performed based on the preset attenuation coefficient and the time difference to obtain the time attenuation factor corresponding to the historical fault mode information. The time attenuation factor decreases as the time difference increases.

[0159] To better align with the principle in industrial quality control that recent high-incidence issues require focused interception while long-term, occasional problems are gradually downgraded, this invention introduces a nonlinear attenuation mathematical model. Specifically, in the background calculation logic, the calculated time difference is combined with a preset attenuation coefficient, and a negative exponential operation is performed. The time attenuation factor output by this operation, constrained by the monotonically decreasing characteristic of the negative exponential function, exhibits a nonlinear attenuation trend as the time difference increases.

[0160] In practical implementation, in order to establish a more suitable quality risk quantification assessment model, the above parameters are integrated into the final risk assessment. The weighted calculation of the risk weight value RPN corresponding to the atomized detection command can be expressed as follows: First, regarding the aforementioned severity parameters Discrete severity rating mapping tables can be established; for example, their value set can be configured as follows. ∈{1(minor),2(moderate),3(serious),4(fatal)}, and then the corresponding score can be directly matched according to the consequences of the failure.

[0161] Secondly, regarding the aforementioned frequency parameters This can be achieved through proportional calculation. For example, by reading from a database, the calculation formula can be set as follows: =Number of failures / Production output in the same period, thus obtaining an objective probability value of occurrence.

[0162] Furthermore, the time decay factor calculated for this embodiment... The specific mathematical expression for the above negative exponentiation operation is: ; in, It is a natural constant. The preset attenuation coefficient, This is the time difference. According to this expression, it represents the time difference corresponding to a recently occurring fault. Smaller The value is closer to 1, thus giving higher weight to recent faults; while the weight of older faults declines exponentially over time.

[0163] Finally, the risk weight values ​​are derived using a comprehensive multiplication model. That is, to execute the calculation formula: ; The resulting RPN final value will be used as the core selection criterion for arranging the aforementioned atomic detection instructions in sequence.

[0164] The equipment performance testing method provided in this invention employs a nonlinear time-effect decay algorithm that ensures that batch quality problems that have just emerged in the market receive amplified weight in a very short time. Combined with the joint modeling of severity classification and objective occurrence frequency, it provides a precise and highly consistent decision-making method for the dynamic adaptive adjustment of the testing strategy, which closely matches the actual quality fluctuation cycle.

[0165] Combination Figure 5 As shown, to address the problem that unstructured quality feedback information is difficult to directly apply to the underlying control of the production line, this embodiment of the invention provides a specific implementation method for quality knowledge extraction based on a large text processing model. The step of obtaining historical fault mode information associated with each atomized detection instruction in a preset instruction library may specifically include the following steps: First, acquire multi-source quality feedback data, which includes at least one of the following: return / exchange work order data, maintenance record text data, and batch quality event data.

[0166] In the initial stage of establishing the automatic transmission link, this embodiment will construct a multi-source data access layer that broadly covers both the marketing and sales ends. In specific implementation, it can interface with various internal information systems of the enterprise to extract feedback sources in different formats.

[0167] Specifically, the data for return and exchange work orders can come from a customer relationship management system and include structured field information such as product model, fault code, reason for return or exchange, and time of occurrence.

[0168] Repair record text data can come from work order attachments uploaded by after-sales engineers. It is mostly unstructured natural language descriptions, such as engineers' on-site fault descriptions, repair logs, or users' original feedback scripts.

[0169] Batch quality event data can come from tag data generated by the quality monitoring platform, such as batch quality problem markings triggered by the system or data related to product recall events.

[0170] This embodiment integrates the above data types to form a comprehensive multi-source quality feedback data base.

[0171] Furthermore, the multi-source quality feedback data is cleaned to obtain cleaned quality feedback data.

[0172] Considering that the raw multi-source data obtained directly often contains a lot of noise, repetitive and redundant content or non-standard descriptions, this embodiment can perform preset data preprocessing before feeding it into the model.

[0173] In practical implementation, natural language processing rule scripts can be used to perform deduplication and invalid character removal operations on the multi-source quality feedback data. Additionally, terminology normalization can be introduced, for example, non-standard colloquial expressions such as "water won't drain" or "water stuck" casually written by engineers can be uniformly mapped and converted into standard engineering terms such as "drainage blockage," thereby minimizing semantic discrepancies and outputting cleaned quality feedback data with a clear structure and low noise.

[0174] Further, the cleaned quality feedback data is input into the text processing model to obtain multiple candidate fault mode information and the confidence score corresponding to each candidate fault mode information after the text processing model sequentially performs entity recognition processing and relation extraction processing. After completing the above data preprocessing, this embodiment will send the cleaned quality feedback data to the text processing model deployed in the background. The text processing model can be a pre-trained Large Language Model (LLM) fine-tuned from the manufacturing quality corpus. This text processing model automatically executes the following two core processing subtasks in its internal pipeline: On the one hand, it is used to perform entity recognition processing, which is used to accurately identify key physical entities and feature words from the quality feedback data after cleaning. For example, through sequence labeling algorithms, product model entities (such as "pulsator model X-1"), component entities (such as "drain valve" and "main control board"), and fault phenomenon entities (such as "no power" and "abnormal washing noise") can be accurately segmented and extracted from large segments of industrial text.

[0175] On the other hand, it can be used to perform relation extraction processing, that is, to establish logical relationships between the product model entity, the component entity, and the fault phenomenon entity based on the captured discrete entities. For example, through a deep understanding of the context, it is possible to construct an accurate mapping relationship between faults and components. For instance, it can determine that the abnormal noise is caused by mechanical jamming of the drain valve, rather than a fault in the main control board, and then assemble and piece together the scattered entities to form a structured expression of the fault mode.

[0176] Based on the entity recognition and execution relationship extraction described above, the text processing model outputs a large amount of structured candidate fault mode information. Simultaneously, the underlying classification network of the text processing model can assign a confidence score from 0 to 1 to each candidate fault mode based on the probability distribution of semantic matching, thereby quantifying the reliability of that candidate fault mode.

[0177] Finally, the candidate fault mode information associated with the atomization detection command among the candidate fault mode information whose confidence scores meet the preset confidence conditions is used as the historical fault mode information.

[0178] Furthermore, this embodiment sets an empirical value as a pre-set confidence condition, such as a confidence threshold of 0.85, to filter all output results, eliminating noisy data that the model is unsure about due to semantic ambiguity, and retaining only highly reliable results. The selected high-quality candidate fault mode information is then correlated with predefined underlying code in a knowledge graph or instruction mapping table. Text information that has passed the above filtering and can be successfully mapped to underlying execution actions is formally confirmed as historical fault mode information associated with a specific atomized detection instruction, and subsequently written into the database for use in the weighted calculation of risk weight values ​​in the aforementioned steps.

[0179] Combination Figure 5 As shown, Figure 5 The outer dashed box indicates the overall processing scope of the text processing pipeline, which, in sequence, includes text cleaning (data cleaning and normalization), entity recognition processing, relation extraction processing, and the output of candidate fault mode information and confidence scores. The inner dashed box further indicates the model boundary of the text processing model; entity recognition processing, relation extraction processing, and confidence score output are performed by the text processing model, while text cleaning is a data preprocessing step performed before multi-source quality feedback data is fed into the text processing model. Confidence filtering is a post-processing step on the output of the text processing model.

[0180] Compared to the traditional, high-latency approach that relies on quality engineers to manually review maintenance reports and summarize fault statistics, this implementation fully leverages the entity recognition and relation extraction capabilities of a large language model. It continuously and automatically extracts and quantifies objective failure characteristics and related components from massive and complex after-sales text. This data pipeline based on a text processing large model opens up a channel for market failure performance to indirectly empower production line detection strategies. This allows early, hidden defects exposed when new products first enter the market to be quickly transformed into highly confident preventative evidence for the production line after semantic extraction.

[0181] Based on the above embodiments, this embodiment of the invention, after associating and storing the risk weight value with the atomization detection instruction, continues execution. Figure 3 The historical learning mechanism shown in the dynamic adaptive adjustment section mainly includes, but is not limited to: Determine whether the preset risk weight value update triggering conditions are met. The risk weight value update triggering conditions include timed triggering conditions and / or event triggering conditions. The timed triggering condition is that the time interval between the current time and the last time the risk weight value update was performed reaches a preset update cycle. The event triggering condition is that a new batch quality event that meets the preset event level is detected in the batch quality event data. If the timed triggering condition or the event triggering condition is met, the risk weight value corresponding to each atomic detection instruction in the candidate detection instruction set is recalculated.

[0182] In one optional implementation, a resident monitoring daemon runs in the background to listen in real time and determine whether the current system status has triggered a pre-defined weight refresh threshold. To balance routine maintenance with emergency response to unforeseen events, this embodiment of the invention configures corresponding triggering conditions: First, the timed trigger condition is configured such that the time interval between the current moment and the last time the risk weight value was updated reaches a preset update cycle.

[0183] During use, the internal timer continuously calculates the elapsed time. When this time interval accumulates to a preset update cycle, such as setting it to early morning every day or weekend every week, the timed trigger condition is determined to be met. This periodic timed trigger mechanism ensures that the large model automatically performs data analysis and knowledge base reconstruction within the preset cycle, achieving a smooth transition of the production line inspection focus with the fluctuations of regular market data.

[0184] In addition, the event triggering condition is set to detect the addition of a batch quality event that meets the preset event level in the batch quality event data.

[0185] In certain extreme cases, such as when a batch of new products experiences widespread, specific malfunctions due to design flaws or poor manufacturing processes shortly after market launch, conventional time-based cycles become too slow. In such situations, the system connects in real-time to the factory's quality monitoring platform. Once a sudden batch quality alert is detected within the multi-source quality feedback data, and the severity or scope of the event meets a predefined event level, the event trigger condition is deemed met, enabling a second-level response to major unforeseen risks.

[0186] Through the above implementation methods, this embodiment significantly shortens the detection strategy update cycle. It changes the existing technology's one- to two-week-long delayed response process for discovering new product defects, which involves manual statistics, request submissions, modification of electronic control code, and recompilation and reprogramming. This embodiment compresses the adjustment cycle of the detection logic to the hourly or even minute level, allowing real-time failure feedback from the market to drive and automatically adjust the resource allocation direction on the production line. This adaptive dynamic update closed loop completely reverses the passive and lagging nature of the traditional detection process, ensuring that limited detection resources are always prioritized for the most critical projects at the earliest possible time.

[0187] Figure 6 This is a schematic diagram of the atomized detection instruction decoupling architecture provided by the present invention, as shown below. Figure 6 As shown, this embodiment of the invention further decomposes the underlying detection code into a structured form, breaking down the original continuous detection program into independent and controllable minimum execution units, thereby designing a standardized atomic instruction set. In specific implementation, the atomic detection instruction includes one or more of the following: instruction identifier field, input parameter field, output result field, and dependency relationship field.

[0188] Combination Figure 6 As shown, the dashed box on the left illustrates the coupling mode used in the prior art, namely, the fixed continuous detection program is written into the electronic control firmware in a hard-coded manner, compiled and burned to the electronic control board, and then the fixed detection process is executed on the production line. Figure 6 The dashed box on the right illustrates the decoupling mode adopted in this invention, which includes an atomic detection instruction set consisting of multiple independently controllable atomic detection instructions such as instruction A, instruction B, instruction C, and instruction D. Each atomic detection instruction includes one or more of the following: instruction identifier field, input parameter field, output result field, and dependency relationship field. The atomic detection instruction set is dynamically combined and then sent to a general instruction execution engine deployed on the target device's electronic control terminal. The general instruction execution engine includes a communication protocol parsing module, an instruction scheduling module, a status monitoring module, and a result feedback module, thereby controlling the target device to dynamically combine and execute different detection processes.

[0189] The instruction identifier field is used to uniquely identify the atomization detection instruction, and the instruction identifier field contains the product category code and the function type code.

[0190] The input parameter field is used to store the standardized input parameters required when the atomization detection instruction is executed.

[0191] The output result field is used to store the standardized output result fed back after the atomization detection instruction is executed.

[0192] The dependency field is used to store the execution association between the atomic detection instruction and other atomic detection instructions.

[0193] In practical implementation, the instruction identifier field can be configured as a preset length of code, such as a 16-bit unique ID. By dividing this encoding structure into specific bit segments, each segment carries a product category code to distinguish the target device's macro-level classification (e.g., refrigerators, washing machines, or air conditioners) and a function type code to characterize specific hardware driving actions (e.g., water inlet, drainage, heating, or motor operation). This encoding mechanism ensures that each smallest execution unit, disassembled from the general-purpose manufacturing and testing platform, possesses a globally unique and precisely addressable identity.

[0194] Considering that each atomization detection instruction is designed as a general logic template that is independent of a specific product model, this embodiment sets specific execution boundaries before it is actually called and executed. The input parameter field serves as a standardized interface for receiving these configuration parameters, and is used to load and store control parameters such as target operating speed, safety limit water level, set operating time upper limit or specific voltage threshold, so as to guide the underlying driver module to carry out testing according to the input standardized parameters.

[0195] After the underlying hardware of the target device executes the current atomization detection instruction, the output result field is used to uniformly represent the various physical characteristics and status information of the output. This output result field is used to standardize the data packets returned by the device microcontroller or sensor network. It can store Boolean success or failure status codes, or it can store specific sensor readings, such as real-time phase current values, speed rise curves, or temperature readings, thereby providing a standardized data source for the aforementioned dynamic adaptive adjustment mechanism that triggers deep detection based on execution result data.

[0196] This dependency field, as a structure record area, internally contains the preconditions that the current atomization detection instruction must meet before it is executed. For example, a "maximum speed dehydration" instruction must depend on the "drainage completed" instruction as a prerequisite, as well as mutually exclusive conditions that cannot be executed simultaneously. The association rules recorded in the above field directly provide underlying physical rule support for the legality verification of the aforementioned dynamic scheduling and screening based on constraints.

[0197] Through the standardized field design described above, this embodiment of the invention decouples the hard-coded coupling between the detection logic and the product's firmware. This not only transforms the complex detection process into an independent module that can be flexibly scheduled by the upper-layer algorithm, effectively simplifying the problem of parallel maintenance of multiple models and versions of the detection program, but also eliminates the need for the lengthy cycle of code modification, firmware compilation, and production line burning when adding, deleting, or adjusting detection items. This greatly improves the agility of industrial production lines in responding to iterative quality risks.

[0198] Based on the above embodiments, combined with Figure 6 The diagram illustrates an atomized detection instruction decoupling architecture. In this embodiment, a lightweight general-purpose instruction execution engine is further deployed on the electronic control side of the target device. In a specific implementation, the target device pre-deploys the general-purpose instruction execution engine, which includes one or more of the following: a communication protocol parsing module, an instruction scheduling module, a status monitoring module, and a result feedback module.

[0199] Correspondingly, sending the target detection instruction sequence to the target device to control the target device to execute the target detection instruction sequence may specifically include the following steps: First, the target detection instruction sequence issued by the host computer scheduling system is received through the communication protocol parsing module, and the target detection instruction sequence is parsed to obtain the parsed target detection instruction sequence.

[0200] Considering the reality of multiple models of equipment coexisting and diverse communication protocols on industrial production lines, the communication protocol parsing module serves as the unified data entry point for the execution engine. During use, it receives standardized data frames issued by the host computer scheduling system in real time through a pre-adapted communication bus, such as a controller local area network bus or a serial communication bus. It then unpacks, verifies, and restores the encapsulated data of the communication protocol layer, restoring the original communication message into an instruction object that can be recognized within the electronic control engine, thereby obtaining the parsed target detection instruction sequence.

[0201] Furthermore, the instruction scheduling module distributes each target atomization detection instruction to the corresponding functional module driver unit in the target device in sequence according to the arrangement order of the parsed target detection instruction sequence, so as to drive the target device to execute the detection condition corresponding to the target atomization detection instruction.

[0202] The instruction scheduling module acts as the central hub for instruction distribution. Internally, it maintains a first-in, first-out instruction execution queue. During scheduling, it generally follows the order output by the aforementioned risk weight sorting and constraint solving. It sequentially retrieves the currently pending target atomization detection instructions from the head of the queue and, based on the function type code in the instruction identifier field, precisely routes them to the corresponding functional module driver, such as a water circuit control driver, motor control driver, or heating control driver. Upon receiving the instruction object and input parameters, each functional module driver further operates the underlying electrical components, thereby driving the target device to enter and execute the specific physical detection condition corresponding to the target atomization detection instruction.

[0203] Furthermore, the status monitoring module monitors the process of the target device executing the detection condition, and obtains the execution result data corresponding to the target atomized detection command.

[0204] Specifically, the status monitoring module continuously collects the physical parameters and operating status codes output by various underlying sensors during the detection process by periodically polling or interrupting.

[0205] Optionally, the status monitoring module can also handle anomalies. If an abnormal event such as a hardware over-limit alarm or communication link interruption is detected during monitoring, the circuit breaker mechanism can be immediately triggered to prevent equipment damage. Once the detection process is completed normally, the status monitoring module summarizes the key parameters collected throughout the process with the final conclusions to form complete execution result data.

[0206] Finally, the execution result data is sent back to the host computer scheduling system via the result feedback module.

[0207] Specifically, the result feedback module encapsulates and packages the execution result data according to the communication protocol specification symmetrically configured with the communication protocol parsing module, and then transmits it back to the host computer scheduling system via the aforementioned communication bus. Upon receiving the execution result data, the host computer scheduling system can either use it to drive the aforementioned dynamic adjustment mechanism that inserts deep detection instructions based on abnormal triggering conditions in real time, or it can write it into a historical database as quality big data for subsequent iterative update calculations of risk weight values.

[0208] This invention, through the decoupling and collaborative operation of four modules—communication protocol parsing, instruction scheduling, status monitoring, and result feedback—eliminates the need for embedded specific detection condition combination logic in the electronic control firmware. It only requires the general capability to understand atomic detection instructions and drive hardware operation. This decoupled execution architecture significantly reduces the number of electronic control firmware versions for home appliances. Simultaneously, the host computer can flexibly control the same target device to execute differentiated detection processes by issuing different instruction combinations, fundamentally resolving the technical bottlenecks of complex version management and cross-model reuse difficulties caused by the deep coupling of detection logic and electronic control firmware in traditional models.

[0209] To more clearly illustrate the overall implementation process of the intelligent agent method for home appliance performance testing provided by this invention in a real-world industrial production line scenario, and the collaborative relationship between each step, the following is a combination of... Figure 7 As shown, the method of the present invention is described in its entirety, taking the production line performance testing of washing machine products as an application scenario.

[0210] like Figure 7 As shown, the home appliance performance testing method provided in this embodiment of the invention includes two mutually cooperating and parallel workflows in practical applications: a market quality data analysis workflow running in the background and a production line testing execution workflow running in the foreground.

[0211] For the market quality data analysis stream, the host computer scheduling system retrieves market return records for different product models from the enterprise customer relationship management system within a preset update cycle (e.g., the past year or two). These market return records are used as a core component of the aforementioned multi-source quality feedback data. Subsequently, the system extracts textual descriptions of the return reasons, entered in natural language by after-sales engineers, from these records. After text cleaning, the cleaned textual descriptions are input into the deployed text processing model. The model automatically performs semantic recognition of the return reason textual descriptions using entity recognition and relation extraction methods provided in the aforementioned embodiments, and accurately matches them with pre-determined atomized detection instructions based on the mapping relationships of a knowledge graph.

[0212] Finally, the host computer scheduling system summarizes the matching output results and outputs the risk weight value corresponding to each detection item according to the aforementioned weighted calculation method based on severity parameters, occurrence frequency parameters, and time decay factors. Each detection item will be sorted according to the risk weight value and stored in the production line detection knowledge base.

[0213] Furthermore, after completing this round of storage, the host computer scheduling system enters a dormant state for a fixed period of time. When the dormant period reaches the preset update cycle, the above-mentioned retrieval and analysis process is repeated again, thereby forming a continuous self-circulating closed loop for updating market quality knowledge.

[0214] For the production line inspection execution flow, when a washing machine is operated as the target device to the inspection station, the station barcode scanner first completes the product arrival scanning action and sends the scanned product SN code back to the host computer scheduling system.

[0215] The SN code obtained by scanning is parsed to obtain information such as the product model and production batch of the target equipment. Based on the parsed product model information, the candidate detection instruction set associated with the model and its respective risk weight value are retrieved from the aforementioned production line detection knowledge base.

[0216] Next, following the aforementioned dynamic programming algorithm for detection sequences based on multiple constraints, the performance detection item planning is executed by comprehensively considering the upper limit of the current workstation's cycle time, the execution correlation between each detection item, and the risk weight value of each detection item. The specific planning process includes steps such as sorting by risk weight value in descending order, filtering by the greedy algorithm's main loop, and filling in remaining time fragments, thereby outputting the final target detection instruction sequence.

[0217] The detection items can be sequentially combined and sent to the washing machine's control board via a pre-defined communication bus. Upon receiving the sequential combination of detection items, the pre-deployed general instruction execution engine on the control board, through the coordinated operation of the communication protocol parsing module, instruction scheduling module, status monitoring module, and result feedback module, respectively instructs and controls the washing machine to operate the specific conditions corresponding to each detection item, such as water inlet, washing motor operation, heating, drainage, and spin-drying. During execution, the general instruction execution engine can also send the real-time execution result data of each detection item back to the host computer scheduling system, and trigger the aforementioned dynamic insertion mechanism of deep detection instructions in case of an anomaly.

[0218] Figure 8 This is a schematic diagram of the structure of the equipment performance testing device provided by the present invention, as shown below. Figure 8 As shown, it mainly includes, but is not limited to: The cycle time acquisition module 21 is used to acquire the upper limit of the cycle time of the production line where the target equipment is located; The instruction acquisition module 22 is used to acquire a candidate detection instruction set, which includes multiple independent atomic detection instructions. Each atomic detection instruction is configured with a risk weight value, execution time, and execution correlation. The sequence generation module 23 is used to maximize the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence. The instruction issuing module 24 is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0219] It should be noted that the device performance testing apparatus provided by the present invention can implement the device performance testing method provided in any of the above embodiments when it is executed, and will not be described in detail here.

[0220] The device performance testing apparatus provided by this invention decouples the testing commands atomically and generates testing sequences based on multi-constraint dynamic programming, thereby decoupling the testing logic from the electronic control firmware, supporting rapid iteration of the testing strategy, and maximizing the quality risk coverage within a limited cycle time.

[0221] Figure 9 This is a schematic diagram of the structure of the intelligent agent for device performance detection provided by the present invention, as shown below. Figure 9 As shown, it mainly includes, but is not limited to: The data processing layer is used to obtain the upper limit of the cycle time issued on the production line where the target device is located; and to obtain a candidate detection instruction set, which contains multiple independent atomic detection instructions, each of which is configured with a risk weight value, execution time and execution correlation. The knowledge reasoning layer is used to maximize the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence. An execution feedback layer is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0222] In specific implementation, combined with Figure 9 As shown, the data processing layer further includes a market quality data access module, a large model text processing module, and a quality risk assessment engine.

[0223] The market quality data access module is used to connect to data sources such as the enterprise's internal customer relationship management system, after-sales maintenance platform, and quality monitoring platform, and to obtain multi-source quality feedback data such as return and replacement work order data, repair record text data, and batch quality event data in accordance with the method described in the aforementioned embodiments.

[0224] The large model text processing module is used to perform structured extraction of the multi-source quality feedback data according to the text cleaning, entity recognition and relation extraction pipeline described in the foregoing embodiments, and output candidate fault mode information with confidence scores.

[0225] The quality risk assessment engine is used to calculate and refresh the risk weight value for each atomized detection command according to the aforementioned weighted calculation method based on severity parameters, occurrence frequency parameters, and time decay factors.

[0226] Meanwhile, the data processing layer can also obtain the upper limit of the current production line cycle time through the interface of the factory manufacturing execution system.

[0227] The knowledge reasoning layer further includes an FMEA knowledge graph engine, an intelligent retrieval module for detection items, and a dynamic programming module for detection sequences.

[0228] The FMEA knowledge graph engine is used to carry the pre-built knowledge graph described in the aforementioned embodiments and to store the relationships between multi-level entities such as product models, functional modules, components, failure modes, testing conditions, and atomic instructions.

[0229] The intelligent retrieval module for detection items is used to retrieve a set of candidate detection instructions associated with the target device from the knowledge graph using the aforementioned multi-dimensional retrieval methods, such as exact matching retrieval, similarity reasoning based on embedded vectors, and transitive reasoning.

[0230] The detection sequence dynamic programming module is used to perform operations such as descending sorting, constraint verification, and filling of remaining time fragments according to the aforementioned heuristic planning method based on greedy algorithm, and finally outputs the target detection instruction sequence.

[0231] The execution feedback layer further includes an instruction communication scheduling module, an electronic control terminal execution engine, and a closed loop for detection result analysis.

[0232] The instruction communication scheduling module is used to encapsulate the above target detection instruction sequence into a data frame conforming to the communication protocol specification and send it to the target device through a preset communication bus, such as a CAN bus or a UART bus.

[0233] The aforementioned electronic control terminal execution engine is the general instruction execution engine deployed on the target device's electronic control board as described in the previous embodiment. Internally, through the coordinated operation of the communication protocol parsing module, instruction scheduling module, status monitoring module, and result feedback module, it drives the target device's various functional modules to execute each detection condition in sequence.

[0234] The closed loop for analyzing detection results is used to receive the execution result data returned by the electronic control end execution engine. On the one hand, it is used to trigger the dynamic insertion mechanism of the aforementioned deep detection command in real time during the detection process. On the other hand, it is used to precipitate the detection results as quality data and return them to the data processing layer to drive the risk weight value refresh in subsequent rounds.

[0235] The data processing layer, the knowledge reasoning layer, and the execution feedback layer have a clear collaborative relationship in terms of data flow. The output of the quality risk assessment engine in the data processing layer is transmitted to the detection sequence dynamic planning module in the knowledge reasoning layer through a vertical data channel, providing a basis for dynamically updated risk weight values ​​for sequence planning. The reasoning output of the FMEA knowledge graph engine in the knowledge reasoning layer is transmitted to the instruction communication scheduling module in the execution feedback layer through a vertical data channel, providing a precise product model and instruction binding relationship for instruction issuance. Through the vertical and horizontal data integration of the above three-layer architecture, the equipment performance detection intelligent agent can independently complete the entire closed-loop operation from market data collection, quality knowledge extraction, risk weight quantification, intelligent retrieval of detection items, dynamic planning of detection sequences to underlying instruction issuance and result feedback.

[0236] It should be noted that the device performance detection intelligent agent provided in the embodiments of the present invention can be deployed in pure software form on a factory-level server, or it can be deployed on the edge computing node of the production line in a combination of software and hardware. The modules in the above-mentioned layers can be centrally deployed in the same physical device, or they can be distributed according to the actual load. The embodiments of the present invention do not make specific limitations in this regard.

[0237] The device performance testing intelligent agent, with its three-layer architecture comprising a data processing layer, a knowledge reasoning layer, and an execution feedback layer, unifies and automates the previously scattered functions across different departments such as quality management, process planning, and electronic control execution. Each module has clearly defined responsibilities, standardized interfaces, and can be independently upgraded and iterated. This not only effectively reduces the overall system coupling and maintenance difficulty but also enables home appliance production lines to dynamically optimize testing strategies end-to-end through the agent's self-driven approach when facing rapidly changing market quality feedback. This comprehensively solves the technical problems of existing technologies, such as hard-coded coupling between testing programs and electronic control firmware, long testing strategy adjustment cycles, and excessive reliance on engineers' subjective experience for testing decisions.

[0238] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 10 As shown, the electronic device may include: a processor 110, a communications interface 120, a memory 130, and a communications bus 140, wherein the processor 110, the communications interface 120, and the memory 130 communicate with each other through the communications bus 140. The processor 110 can call logical instructions in the memory 130 to execute a device performance detection method. This method includes: obtaining the upper limit of the cycle time on the production line where the target device is located; obtaining a candidate detection instruction set, the candidate detection instruction set containing multiple independent atomic detection instructions, each of which is configured with a risk weight value, execution time, and execution correlation; selecting multiple target atomic detection instructions from the candidate detection instruction set and arranging them in sequence, with the optimization objective being to maximize the total risk weight value, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time and that each selected target atomic detection instruction satisfies the execution correlation, to generate a target detection instruction sequence; and sending the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0239] Furthermore, the logical instructions in the aforementioned memory 130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0240] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, the computer can execute the equipment performance detection method provided in the above embodiments, the method including: obtaining the upper limit of the cycle time issued on the production line where the target equipment is located; obtaining a candidate detection instruction set, the candidate detection instruction set including multiple independent atomic detection instructions, each of the atomic detection instructions being configured with a risk weight value, execution time and execution correlation; with maximizing the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution correlation, selecting multiple target atomic detection instructions from the candidate detection instruction set and arranging them in sequence to generate a target detection instruction sequence; sending the target detection instruction sequence to the target equipment to control the target equipment to perform detection according to the target detection instruction sequence.

[0241] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the device performance detection method provided in the above embodiments. The method includes: obtaining the upper limit of the cycle time issued on the production line where the target device is located; obtaining a candidate detection instruction set, the candidate detection instruction set containing multiple independent atomic detection instructions, each of the atomic detection instructions being configured with a risk weight value, execution time, and execution correlation; with maximizing the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time and that each selected target atomic detection instruction satisfies the execution correlation, selecting multiple target atomic detection instructions from the candidate detection instruction set and arranging them in sequence to generate a target detection instruction sequence; and sending the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

[0242] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0243] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0244] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of detecting performance of a device, characterized by, include: Obtain the maximum cycle time limit of the production line where the target equipment is located; Obtain a candidate detection instruction set, which contains multiple independent atomic detection instructions, each of which is configured with a risk weight value, execution time, and execution correlation. With the goal of maximizing the total risk weight value, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, multiple target atomic detection instructions are selected from the candidate detection instruction set and arranged in sequence to generate a target detection instruction sequence. The target detection instruction sequence is sent to the target device to control the target device to perform detection according to the target detection instruction sequence.

2. The equipment performance testing method according to claim 1, characterized in that, The step of selecting multiple target atomized detection instructions from the candidate detection instruction set and arranging them in sequence to generate a target detection instruction sequence includes: According to the risk weight values ​​from high to low, the atomic detection instructions in the candidate detection instruction set are sorted to obtain the sorted candidate detection instruction sequence. According to the sorted candidate detection instruction sequence, each of the atomic detection instructions is taken as the current instruction to be evaluated in turn, and it is determined whether the current instruction to be evaluated satisfies the constraint conditions. If the conditions are met, the current instruction to be evaluated is added to the selected instruction set as the target atomization detection instruction; if the conditions are not met, the current instruction to be evaluated is skipped. After determining all the atomic detection instructions in the sorted candidate detection instruction sequence, the target atomic detection instructions in the selected instruction set are arranged in the order of addition to generate the target detection instruction sequence.

3. The equipment performance testing method according to claim 2, characterized in that, The step of determining whether the current instruction to be evaluated satisfies the constraint conditions includes: Initialize the remaining available time to the maximum cycle time. For the current instruction to be evaluated, based on the execution time of the current instruction to be evaluated and the remaining available time, it is determined whether the current instruction to be evaluated meets the constraint conditions; When the instruction to be evaluated is added to the selected instruction set, the execution time of the instruction to be evaluated is subtracted from the remaining available time to obtain the updated remaining available time. The updated remaining available time is used to determine the constraint conditions for the next atomic detection instruction.

4. The equipment performance testing method according to claim 3, characterized in that, The step of determining whether the current instruction to be evaluated satisfies the constraint condition based on the execution time of the current instruction to be evaluated and the remaining available time includes: Determine whether the execution time of the currently evaluated instruction is not greater than the remaining available time to obtain a first determination result; Determine whether the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship to obtain a second determination result; If the first judgment result indicates that the execution time of the current instruction to be evaluated is not greater than the remaining available time, and the second judgment result indicates that the current instruction to be evaluated and each of the target atomization detection instructions in the selected instruction set satisfy the execution association relationship, then the current instruction to be evaluated is determined to satisfy the constraint condition.

5. The equipment performance testing method according to claim 3, characterized in that, After completing the judgment of all the atomic detection instructions in the sorted candidate detection instruction sequence, the method further includes: Determine whether the remaining available time is greater than zero; If the remaining available time is greater than zero, atomic detection instructions that have not been added to the selected instruction set and whose execution time is not greater than the remaining available time are selected from the candidate detection instruction set and are used as the instruction set to be filled. According to the order of execution time from shortest to longest, each atomic detection instruction in the instruction set to be filled is taken as the current instruction to be filled, so as to determine whether the current instruction to be filled satisfies the constraint condition. If the conditions are met, the current instruction to be filled is added to the selected instruction set as the target atomicity detection instruction, and the execution time of the current instruction to be filled is deducted from the remaining available time. If the conditions are not met, the current instruction to be filled is skipped until the remaining available time is insufficient to accommodate the execution time of any atomicity detection instruction in the instruction set to be filled. Arrange the target atomization detection instructions in the selected instruction set after filling according to the order of addition, and update the target detection instruction sequence.

6. The equipment performance testing method according to claim 1, characterized in that, After sending the target detection command sequence to the target device, the method further includes: Receive execution result data fed back by the target device for the target atomization detection instructions that have been executed in the target detection instruction sequence; Determine whether the execution result data meets the preset exception triggering conditions; If the execution result data meets the abnormal triggering condition, based on the execution result data, a depth detection instruction corresponding to the abnormal triggering condition is obtained from a preset depth detection instruction library; The depth detection instruction is inserted after the executed target atomization detection instruction in the target detection instruction sequence; Based on the execution time of the depth detection instruction and the execution time of the unexecuted target atomic detection instructions in the target detection instruction sequence, target atomic detection instructions that meet the preset removal rules are removed from the target detection instruction sequence to obtain an updated target detection instruction sequence. Send the depth detection instruction and the unexecuted target atomization detection instruction from the updated target detection instruction sequence to the target device to control the target device to continue performing detection according to the updated target detection instruction sequence.

7. The equipment performance testing method according to claim 1, characterized in that, The method for obtaining the candidate detection instruction set includes: Obtain fault mode information associated with the target device; Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomized detection instructions corresponding to the fault mode information. The knowledge graph pre-stores the mapping relationship between fault modes and atomized detection instructions. All the retrieved atomized detection instructions are summarized to obtain the candidate detection instruction set.

8. The equipment performance testing method according to claim 7, characterized in that, The knowledge graph pre-stores the mapping relationship between product model information and failure mode information and atomized detection instructions, and the failure mode information includes the target product model information and target failure mode information of the target device. Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: Use the target product model information and the target failure mode information as the joint search key; Search the knowledge graph for product model information and failure mode information that match the joint search key, and use the atomized detection instructions mapped by the found product model information and failure mode information as the atomized detection instructions corresponding to the failure mode information.

9. The equipment performance testing method according to claim 7, characterized in that, The knowledge graph pre-stores multiple known failure mode information and atomic detection instructions corresponding to each known failure mode information. The failure mode information includes the target failure mode information of the target device. Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: Feature extraction is performed on the target failure mode information to obtain the target embedding vector corresponding to the target failure mode information; Obtain the known embedding vectors corresponding to each known failure mode information in the knowledge graph; Calculate the semantic similarity between the target embedding vector and each of the known embedding vectors; The known failure mode information corresponding to the known embedding vector whose semantic similarity satisfies the preset similarity condition is used as the similar failure mode information; The atomized detection instructions mapped to the similar failure mode information in the knowledge graph are used as the atomized detection instructions corresponding to the failure mode information.

10. The equipment performance testing method according to claim 7, characterized in that, The knowledge graph pre-stores fault phenomenon nodes, component nodes, functional module nodes, and atomization detection instruction nodes. The fault phenomenon nodes are associated with the component nodes, the component nodes are associated with the functional module nodes, and the functional module nodes are associated with the atomization detection instruction nodes. The fault mode information includes the target fault phenomenon information of the target device. Based on the fault mode information, a mapping retrieval is performed in a pre-constructed knowledge graph to obtain multiple atomic detection instructions corresponding to the fault mode information, including: In the knowledge graph, identify the target fault phenomenon node that matches the target fault phenomenon information; Based on the association between the target fault phenomenon node and each of the component nodes, the target component node associated with the target fault phenomenon node is determined; Based on the association between the target component node and each of the functional module nodes, the target functional module node associated with the target component node is determined; Based on the association between the target functional module node and each of the atomization detection instruction nodes, the target atomization detection instruction node associated with the target functional module node is determined; The atomic detection instruction corresponding to the target atomic detection instruction node is used as the atomic detection instruction corresponding to the fault mode information.

11. The equipment performance testing method according to claim 1, characterized in that, Before obtaining the candidate detection instruction set, the method further includes: Obtain historical fault mode information associated with each of the atomic detection instructions in the preset instruction library; Obtain the severity parameter, occurrence frequency parameter, and time decay factor corresponding to the historical fault mode information. The severity parameter is used to characterize the severity of the fault consequences corresponding to the historical fault mode information. The occurrence frequency parameter is used to characterize the occurrence frequency of the historical fault mode information within a preset statistical period. The time decay factor is used to characterize the influence weight of the occurrence time of the historical fault mode information on the current moment. The risk weight value corresponding to the atomized detection command is obtained by weighting the severity parameter, the occurrence frequency parameter, and the time decay factor. The risk weight value is associated with and stored in relation to the atomization detection instruction.

12. The equipment performance testing method according to claim 11, characterized in that, Obtaining the time decay factor corresponding to the historical fault mode information includes: Obtain the fault occurrence time corresponding to the historical fault mode information; Calculate the time difference between the current moment and the time when the fault occurred; A negative exponential calculation is performed based on a preset attenuation coefficient and the time difference to obtain the time attenuation factor corresponding to the historical fault mode information. The time attenuation factor decreases as the time difference increases.

13. The equipment performance testing method according to claim 11, characterized in that, The step of obtaining historical fault mode information associated with each of the atomic detection instructions in the preset instruction library includes: Acquire multi-source quality feedback data, which includes at least one of return / exchange work order data, maintenance record text data, and batch quality event data; The multi-source quality feedback data is cleaned to obtain cleaned quality feedback data; The cleaned quality feedback data is input into the text processing model to obtain multiple candidate fault mode information and the confidence score corresponding to each candidate fault mode information after the text processing model performs entity recognition processing and relation extraction processing in sequence. The entity recognition process is used to identify product model entities, component entities, and fault phenomenon entities from the cleaned quality feedback data, and the relationship extraction process is used to establish the association relationship between the product model entities, the component entities, and the fault phenomenon entities. The candidate fault mode information associated with the atomization detection command among the candidate fault mode information whose confidence scores meet the preset confidence conditions is used as the historical fault mode information.

14. The equipment performance testing method according to claim 13, characterized in that, After associating and storing the risk weight value with the atomization detection instruction, the method further includes: Determine whether the preset risk weight value update triggering conditions are met. The risk weight value update triggering conditions include timed triggering conditions and / or event triggering conditions. The timed triggering condition is that the time interval between the current time and the last time the risk weight value update was performed reaches a preset update cycle. The event triggering condition is that a new batch quality event that meets the preset event level is detected in the batch quality event data. If the timed triggering condition or the event triggering condition is met, the risk weight value corresponding to each atomic detection instruction in the candidate detection instruction set is recalculated.

15. The equipment performance testing method according to claim 1, characterized in that, The atomization detection instruction includes one or more of the following: instruction identifier field, input parameter field, output result field, and dependency relationship field; The instruction identifier field is used to uniquely identify the atomization detection instruction, and the instruction identifier field contains a product category code and a function type code; The input parameter field is used to store the standardized input parameters required when the atomization detection instruction is executed; The output result field is used to store the standardized output result fed back after the atomization detection instruction is executed; The dependency field is used to store the execution association between the atomic detection instruction and other atomic detection instructions.

16. The equipment performance testing method according to claim 1, characterized in that, The target device is equipped with a general instruction execution engine, which includes a communication protocol parsing module, an instruction scheduling module, a status monitoring module, and a result feedback module. Sending the target detection command sequence to the target device to control the target device to perform detection according to the target detection command sequence includes: The target detection instruction sequence is sent to the communication protocol parsing module of the target device through a preset communication protocol, so as to control the communication protocol parsing module to perform protocol parsing on the target detection instruction sequence to obtain each target atomization detection instruction; The instruction scheduling module is controlled to schedule each of the target atomization detection instructions to the corresponding functional module driver unit of the target device for execution in the order of the target detection instruction sequence. The status monitoring module is controlled to monitor the execution status of each target atomization detection instruction in real time, and triggers a preset exception handling process when an abnormal execution status is detected. The control result feedback module collects the execution result data of each of the target atomization detection instructions and sends the execution result data back through the communication protocol.

17. A device for testing equipment performance, characterized in that, include: The cycle time acquisition module is used to obtain the upper limit of the cycle time issued by the production line where the target equipment is located; The instruction acquisition module is used to acquire a candidate detection instruction set, which contains multiple independent atomic detection instructions. Each atomic detection instruction is configured with a risk weight value, execution time, and execution correlation. The sequence generation module is used to maximize the total risk weight value as the optimization objective, and to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence, subject to the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time and that each selected target atomic detection instruction satisfies the execution association relationship. The instruction issuing module is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

18. An intelligent agent for equipment performance detection, characterized in that, include: The data processing layer is used to obtain the upper limit of the cycle time of the production line where the target equipment is located; And to obtain a candidate detection instruction set, wherein the candidate detection instruction set contains multiple independent atomic detection instructions, and each atomic detection instruction is configured with a risk weight value, execution time and execution correlation; The knowledge reasoning layer is used to maximize the total risk weight value as the optimization objective, and with the constraints that the total execution time of all selected target atomic detection instructions does not exceed the upper limit of the cycle time, and that each selected target atomic detection instruction satisfies the execution association relationship, to select multiple target atomic detection instructions from the candidate detection instruction set and arrange them in sequence to generate a target detection instruction sequence. An execution feedback layer is used to send the target detection instruction sequence to the target device to control the target device to perform detection according to the target detection instruction sequence.

19. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the device performance testing method as described in any one of claims 1 to 16.

20. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the device performance testing method as described in any one of claims 1 to 16.

Images