A Method and System for Defect Detection on Manufacturing Production Lines Based on Big Data Intelligent Algorithms

Abstract

This invention discloses a method and system for manufacturing line defect detection based on big data intelligent algorithms, belonging to the field of manufacturing line defect detection technology. The method includes: acquiring project production line information and constructing a detection cycle chain; deploying a production line detection array, deploying a first computing node based on event triggering based on prior features, deploying a second computing node based on directional detection inference based on event black boxes, deploying a third computing node based on causal inference based on directional detection, and initializing a memristor with a cross-array of nodes; establishing array communication interaction, performing distributed self-driven defect composite logic detection management, and executing event defect alarms. This invention solves the technical problem in existing technologies where manufacturing line defect detection cannot achieve distributed self-driven detection and real-time alarms, resulting in low defect detection efficiency and insufficient accuracy. It achieves distributed self-driven detection and real-time alarms for manufacturing line defects, improving the efficiency and accuracy of defect detection.

Description

Technical Field

[0001] This invention relates to the field of production line defect detection technology, and specifically to a method and system for detecting manufacturing production line defects based on big data intelligent algorithms. Background Technology

[0002] In the field of manufacturing line defect detection, traditional methods often suffer from low detection efficiency and insufficient accuracy, making it difficult to achieve real-time monitoring and early warning of production line defects. Existing technologies often lack deep knowledge reasoning of production line information, failing to effectively locate the correlation between prior features and defect events. Furthermore, the deployment of detection arrays lacks systematicity, making it difficult to form a distributed, self-driven detection mechanism. Additionally, the lack of efficient storage and computing architecture hinders real-time processing and analysis of detection data, making it difficult to meet the high-precision and high-efficiency requirements of modern intelligent manufacturing for production line defect detection.

[0003] Existing technologies suffer from the technical problem that they cannot achieve distributed self-driven detection and real-time alarms for defect detection on manufacturing production lines, resulting in low efficiency and insufficient accuracy in defect detection. Summary of the Invention

[0004] This application provides a manufacturing production line defect detection method and system based on big data intelligent algorithms, which is used to address the technical problems of low efficiency and insufficient accuracy in defect detection in existing manufacturing production lines, which cannot achieve distributed self-driven detection and real-time alarm.

[0005] In view of the above problems, this application provides a method and system for detecting defects in manufacturing production lines based on big data intelligent algorithms.

[0006] The first aspect of this application provides a method for detecting defects on a manufacturing production line based on big data intelligent algorithms, the method comprising:

[0007] The system acquires project production line information, locates prior features and defect events through knowledge reasoning, and constructs a detection cycle chain. Based on this detection cycle chain, it deploys a production line detection array. A first computing node is deployed based on event triggering using prior features, a second computing node is deployed based on directional detection reasoning using event black boxes, and a third computing node is deployed based on causal reasoning using directional detection. Memristors are initialized using a cross-array of nodes as composite memristors, where the initialization criteria include storage and computing areas. The system establishes array communication interaction between the production line detection array and the composite memristors, and performs distributed, self-driven defect composite logic detection management as the project production progresses, executing event defect alarms.

[0008] A second aspect of this application provides a manufacturing line defect detection system based on big data intelligent algorithms, the system comprising:

[0009] The detection cycle chain construction module is used to acquire project production line information, locate prior features-defect events through knowledge reasoning, and construct the detection cycle chain. The production line detection array deployment module is used to deploy the production line detection array according to the detection cycle chain. It deploys the first computing node based on event triggering based on prior features, deploys the second computing node based on directional detection reasoning based on event black box, and deploys the third computing node based on causal reasoning based on directional detection. It initializes memristors with the cross array of nodes as composite memristors, wherein the initialization standard includes storage area and computing area. The event defect alarm module is used to establish array communication interaction between the production line detection array and the composite memristor, perform distributed self-driven defect composite logic detection management as the project production progresses, and execute event defect alarms.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The system acquires project production line information, locates prior features and defect events through knowledge reasoning, and constructs a detection cycle chain. Based on this cycle chain, a production line detection array is deployed. A first computing node is deployed based on event triggering using prior features; a second computing node is deployed based on directional detection reasoning using event black boxes; and a third computing node is deployed based on causal reasoning using directional detection. Memristors are initialized using a cross-array of nodes, serving as composite memristors. Array communication is established between the production line detection array and the composite memristors. As the project production progresses, distributed, self-driven defect composite logic detection management is implemented, and event defect alarms are executed. This achieves distributed, self-driven detection and real-time alarming of manufacturing line defects, improving the efficiency and accuracy of defect detection. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0013] Figure 1 A schematic diagram of the manufacturing line defect detection method based on big data intelligent algorithms provided in this application embodiment;

[0014] Figure 2 A schematic diagram of the structure of a manufacturing line defect detection system based on big data intelligent algorithms provided in this application embodiment.

[0015] Figure labeling: Detection cycle chain construction module 10, production line detection array deployment module 20, event defect alarm module 30. Detailed Implementation

[0016] This application provides a manufacturing production line defect detection method and system based on big data intelligent algorithms, which is used to address the technical problems of low efficiency and insufficient accuracy in existing manufacturing production line defect detection technologies, which cannot achieve distributed self-driven detection and real-time alarm.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a manufacturing production line defect detection method based on big data intelligent algorithms, the method comprising:

[0019] Step S100: Obtain project production line information, locate prior features-defect events through knowledge reasoning, and construct a detection cycle chain.

[0020] Specifically, a sensor network (such as vibration sensors and temperature sensors) deployed at key workstations on the production line collects real-time production line information, including equipment operating parameters and process data, and transmits the data to a central data processing unit. Knowledge graph technology is used to perform knowledge reasoning on the collected data, establishing a mapping between historical defect events and corresponding prior features (such as abnormal equipment vibration frequencies and temperature mutation thresholds), forming a prior feature-defect event knowledge base. Based on this knowledge base, according to the production line flow and process sequence, the temporal relationship between possible prior features and defect events at each stage is identified, constructing a detection cycle chain with time-series constraints. This clarifies the prior features to be monitored and the expected defect events within each detection cycle, providing a basis for subsequent detection array deployment and defect detection.

[0021] Step S200: Deploy the production line detection array according to the detection cycle chain, deploy the first computing node based on event triggering based on prior features, deploy the second computing node based on directional detection inference based on event black box, deploy the third computing node based on causal inference based on directional detection, and initialize the memristor with the cross array of the nodes as a composite memristor, wherein the initialization criteria include the storage area and the computing area.

[0022] Specifically, based on the constructed detection cycle chain, guided by detection features, any node in the detection cycle chain is first identified as the first detection node. For this node, the corresponding prior features and defect events are determined. Then, the detection type based on the prior features (such as physical quantity detection like temperature and pressure) and the production line detection location (such as key equipment components) are defined and set as a Class I code. Simultaneously, the detection type based on the defect events (such as abnormal vibration and noise) and the production line detection location are determined and set as a Class II code. Following this, a front-end deployment including a precursor detection array and an event detection array is executed. Based on this, a first computing node is deployed using prior feature verification combined with event probability inference for event-triggered detection. For each defect event, a one-to-one corresponding event causal chain is mined, and this chain is reversed... The second computing node is deployed by reverse-engineering the detection features of defect events (because the time limit for detecting risk events in dynamic production lines is tight, the black box principle is adopted to improve detection efficiency). The chain is then forwarded to deploy the third computing node by using the detection features to infer the event. Finally, the first computing node array is arranged to form the first composite layer (each node corresponds to a prior feature). The second and third computing node arrays are arranged and cascaded to form the second composite layer (triggered by the first composite layer). After cascading the two layers, the memristor computing area is initialized. At the same time, the storage area, which includes the library storage sub-area and the real-time storage sub-area, is initialized by distributed array arrangement at the node level. A cross-cascaded mapping between the storage area and the computing area is established, thereby forming a composite memristor that supports in-memory computing. This standardizes the detection system and avoids redundant detection.

[0023] Step S300: Establish array communication interaction between the production line detection array and the composite memristor, and perform distributed self-driven defect composite logic detection management as the project production progresses, and execute event defect alarms.

[0024] Specifically, an array communication interaction mechanism is established between the production line detection array and the composite memristor, enabling the data collected by the detection array to be transmitted in real time to the storage and computing areas of the composite memristor. As the production line operates, the first computing node of the first composite layer is triggered to detect prior feature data and transmit it back to the real-time storage sub-area. Simultaneously, an event trigger probability determination is performed: if the prior result does not trigger, the detection thread is terminated; if triggered, based on the second computing node of the second composite layer, the event features are deduced from the prior event, driving the detection equipment to perform directional detection. After the event data is transmitted back to storage, the event is deduced through the third computing node to determine the detection result. When a fault event is detected in the event detection result, an alarm message is generated based on the fault status, the corresponding event detection array is located to execute a front-end alarm response, and an alarm is simultaneously displayed on the mobile terminal via an interface pop-up, realizing a distributed, self-driven defect composite logic detection management and event defect alarm process.

[0025] In one possible implementation, step S200 further includes:

[0026] Step S210: Identify a first detection node according to the detection cycle chain, wherein the first detection node is any node in the detection cycle chain.

[0027] Step S220: For the first detection node, determine the first prior feature and the first defect event.

[0028] Step S230: Determine the detection type and production line detection location based on the first prior feature, and set them as a type of code.

[0029] Step S240: Determine the detection type and production line detection location based on the first defect event, and set it as a Class II code.

[0030] Step S250: Execute the front-end deployment of the first production line detection array according to the first type of coding and the second type of coding.

[0031] Step S260: Wherein, the corresponding prior features and the defect event satisfy a preset occurrence interval, and the production line detection array includes a precursor detection array and an event detection array.

[0032] Specifically, based on the established detection cycle chain, any node in the chain is selected as the first detection node. This node serves as the basic unit for production line defect detection and can be selected according to the time sequence or logical relationship of the production process to meet the needs of different detection scenarios and stages, laying the foundation for subsequent analysis of prior features and defect events for this node.

[0033] For the first identified detection node, by analyzing the operating parameters of the production line equipment, historical production data, and process knowledge, the first prior features related to the node are mined from a large amount of data. These include specific temperature fluctuation ranges, abnormal pressure thresholds, and vibration frequency changes during equipment operation, which are early signs of defects. At the same time, by combining the production line's past defect records and failure cases, the first defect event corresponding to the first prior feature is determined, such as specific defect situations like equipment component wear, circuit short circuits, and functional abnormalities. The correlation between the prior features and defect events is clarified, providing an accurate basis for subsequent detection type determination and array deployment.

[0034] After identifying the first prior feature corresponding to the first detection node, the appropriate detection type is determined based on the physical properties and process characteristics of this prior feature. For example, if the prior feature is abnormal bearing temperature, the detection type is set to temperature sensing detection; if it is abnormal current fluctuation, it is set to electrical parameter detection. Simultaneously, considering the production line layout and equipment structure, the production line detection location where the first prior feature is most likely to manifest is located. For example, bearing temperature detection corresponds to the bearing mounting location, and current detection corresponds to the circuit interface. Subsequently, the detection type and location information are standardized and encoded to form a code. This code uniquely identifies the detection requirement based on the prior feature, providing precise parameter basis for the subsequent front-end deployment of the production line detection array.

[0035] After identifying the first defect event corresponding to the first detection node, the appropriate detection type is determined based on the manifestation and mechanism of the defect event. For example, if the first defect event is wear on equipment components, the detection type is set to visual image detection (capturing wear marks on components through a camera) or vibration spectrum analysis (detecting abnormal vibrations through sensors); if it is a functional failure, it is set to performance index testing (such as pressure, flow rate, and other parameter detection). Simultaneously, combining the production line process flow and equipment layout, the production line detection locations where the first defect event is most likely to occur or most directly reflected are precisely located. For example, wear detection corresponds to the exposed interfaces of moving parts of the equipment, and functional failure detection corresponds to the sensor installation locations at key process nodes. Subsequently, the detection type and location information are structured and encoded to form a second-class code. This code is specifically used to identify the detection requirements based on defect events, complementing the first-class code, and together providing complete parameter specifications for the front-end deployment of the production line detection array, ensuring that the detection equipment can accurately capture real-time data of defect events.

[0036] Based on the pre-defined detection types and production line detection location information contained in the first and second type codes, a first production line detection array is installed and deployed at the corresponding locations on the production line. The first type code corresponds to detection requirements based on first prior features, requiring the deployment of early warning detection devices such as temperature sensors and vibration monitors at designated locations to capture early characteristic signals before defects occur. The second type code corresponds to detection requirements based on first defect events, requiring the deployment of event detection devices such as visual inspection cameras and performance testing instruments at corresponding locations to monitor the occurrence of defect events in real time. By installing and debugging the detection devices corresponding to the two types of codes according to the preset array layout, a front-end detection system including early warning detection arrays and event detection arrays is formed. This enables the detection devices to work collaboratively, achieving accurate detection and data acquisition of relevant prior features and defect events at the first detection node, and ensuring that the detection of prior features and defect events meets the preset occurrence interval requirements, thereby providing reliable front-end data support for subsequent defect complex logic detection management.

[0037] When deploying the first production line inspection array, it is necessary to ensure that the prior features of the corresponding first inspection node and the defect event meet the preset occurrence interval. This interval is set based on the time pattern from the appearance of prior features to the occurrence of defect events in historical production line data to ensure the rationality of the inspection sequence. Simultaneously, the deployed production line inspection array includes a precursor detection array and an event detection array: the precursor detection array is used to capture early characteristic signals before defect occurrence, such as abnormal fluctuations in equipment operating parameters; the event detection array is used to monitor the actual occurrence of defect events in real time, such as physical changes or functional abnormalities during component failure. The two types of arrays are deployed collaboratively to form a detection network covering the entire defect occurrence cycle, achieving both early warning of potential risks and accurate capture of actual defect events, thereby constructing a complete front-end inspection system.

[0038] In one possible implementation, step S200 further includes:

[0039] Step S270: Deploy the first computing node based on prior feature testing and event probability inference.

[0040] Step S280: For defect events, mine the event causal chain, where there is a one-to-one correspondence between the defect event and the event causal chain.

[0041] Step S290: Deploy the second computing node and the third computing node based on event reasoning based on the event causal chain.

[0042] Specifically, the Bayesian classification algorithm from machine learning is used to model the correlation between prior features and defect events in historical production line data. By collecting a large amount of sample data on prior features (such as equipment vibration frequency and temperature thresholds) and their corresponding defect events, the posterior probability of a defect event occurring when a prior feature appears is calculated, thus constructing an event probability inference model. For example, the Naive Bayes algorithm is used, assuming that each prior feature is independent, to calculate the probability of a defect event occurring when a specific combination of prior features is observed, based on the Bayesian formula. The trained model is deployed to the first computing node, which receives prior feature data collected by the precursor detection array in real time. The data is processed using the Bayesian inference algorithm, and when the calculated probability of a defect event exceeds a preset threshold, the event detection process is triggered, realizing event probability inference based on prior features and intelligent deployment of the first computing node.

[0043] For identified defect events, historical production line fault data, equipment operation logs, and process flow diagrams are analyzed. Using methods such as Fault Tree Analysis (FTA) and Root Cause Analysis (RCA), the direct, indirect, and underlying causes of the defect event are traced backward. The subsequent impacts and derivative faults that the event may trigger are deduced forward, constructing a causal chain containing nodes at each level of cause-effect relationships. For example, if the defect event is "bearing overheating and damage," the causal chain can be traced upward to direct causes such as "insufficient lubrication" and "excessive load," and further to underlying factors such as "lubricating oil pump failure" and "sensor malfunction." Simultaneously, potential subsequent events such as "gearbox wear" and "machine shutdown" are deduced downwards. This ensures that each defect event corresponds to a unique and complete causal chain, clearly defining the causes, consequences, and impact paths of the event, providing a precise logical basis for the subsequent deployment of computational nodes based on the causal chain.

[0044] Based on the discovered event causal chains, the deployment of the second and third computing nodes is achieved through event reasoning. The event causal chain is reversed, starting from the defect event and working backward to deduce the required detection features. For example, if the end of the causal chain is "bearing overheating damage," then features such as "lubrication system pressure" and "bearing temperature threshold" need to be detected. The second computing node is deployed accordingly, enabling it to drive the collection and analysis of detection features based on the defect event. The event causal chain is forward-engineered, starting from the detection features and working forward to deduce the event development path. For example, the probability of the "bearing wear" event is deduced from the feature of "abnormal lubrication system pressure." The third computing node is deployed accordingly, enabling it to predict the event evolution trend based on the detection features. Through bidirectional event reasoning, the functional positioning, data processing logic, and deployment parameters of the second and third computing nodes are determined, forming a bidirectional detection reasoning system for defect events.

[0045] In one possible implementation, step S290 further includes:

[0046] Step S291: Reverse the event causal chain, reverse the detection features from the defect events, and deploy the second computing node.

[0047] Step S292: Correct the event causal chain, perform event deduction based on detection features, and deploy the third computing node.

[0048] Specifically, the discovered causal chain of events is reversed. Starting with a defect event, the causal chain is traced back to the various levels of factors that led to the event, thereby deriving detection features that can characterize these factors. For example, if the defect event is a "circuit board short circuit," the reverse causal chain can be traced back to direct causes such as "component aging" and "poor soldering," further leading to detection features such as "abnormal component temperature" and "excessive solder joint resistance." Based on these reverse-derived detection features, the functional modules and data processing flow of the second computing node are determined, and its location on the production line is deployed to collect and analyze these detection features in real time, enabling it to drive the collection and reasoning of detection features from the defect event.

[0049] Maintaining the forward order of the event causal chain, starting with the detected features, the development trend of the event is deduced based on the logical relationships and influence paths of each factor in the causal chain. For example, if the feature "abnormal vibration frequency of motor bearing" is detected, the forward causal chain can be deduced to "increased bearing wear," and further deduction may lead to events such as "motor seizure." Based on this event deduction logic based on detected features, the algorithm model and parameter configuration of the third computing node are determined, and it is deployed in a position on the production line that can receive detected feature data and perform event prediction. This enables it to have the ability to forward predict the evolution and probability of events based on detected features, thereby achieving proactive detection of defect events.

[0050] In one possible implementation, step S200 further includes:

[0051] The first computing nodes are arrayed to determine the first composite layer, wherein each first computing node corresponds to a prior feature; the second and third computing nodes are arrayed and cascaded to determine the second composite layer, wherein the first composite layer triggers the second composite layer; the first and second composite layers are cascaded to initialize the computing region of the memristor.

[0052] Specifically, when arranging the first computing nodes in an array to determine the first composite layer, multiple first computing nodes need to be arranged in a logical order according to the production line's inspection requirements. Each node uniquely corresponds to a priori feature (such as abnormal equipment temperature, vibration frequency fluctuations, etc.). Through this array arrangement, each node can process data of different priori features in parallel, forming a first composite layer with multi-feature synchronous detection capabilities, providing structured computing unit support for subsequent event-triggered detection based on priori features.

[0053] When arranging and cascading the second and third computing nodes to determine the second composite layer, it is necessary to first arrange multiple second computing nodes (used for reverse deduction of defect event detection features) and third computing nodes (used for forward deduction of events from detection features) in an orderly manner according to the data processing flow, forming an array structure, based on the reverse deduction logic and forward deduction requirements of the event causal chain. Then, the two types of nodes are cascaded through a data bus and logic control link, so that the reverse deduction results of the second computing nodes can be used as input parameters for the third computing nodes, constructing a computing network with causal bidirectional reasoning capabilities. Simultaneously, the triggering mechanism of the second composite layer is set so that the operational logic of this layer is activated only after the first composite layer detects a priori features and outputs a trigger signal, ensuring that the detection process follows the temporal logic of "priori feature triggering → event causal reasoning," thus forming a second composite layer driven by the first composite layer, achieving deep reasoning detection of defect events.

[0054] When cascading the first and second composite layers and initializing the memristor computing area, the output port of the first composite layer and the input port of the second composite layer are first physically connected via a high-speed data bus. Simultaneously, a logic control link is established to ensure that the prior feature data processed by the first composite layer can be transmitted to the second composite layer according to the timing rule of "prior feature detection → event causal reasoning". Next, cross-array wiring technology is used to electrically connect the computing nodes of the two layers, forming a matrix signal interaction network between each first computing node (corresponding to a single prior feature) in the first composite layer and the second and third computing nodes (corresponding to event inverse reasoning and forward reasoning) in the second composite layer. During the computing area initialization phase, the resistance state of the memristor array is configured by loading a preset weight parameter table. Pulse voltage signals are used to program the memristor cells, making their resistance values ​​correspond to the causal correlation strength between prior features and defect events. Simultaneously, the logic control unit of the computing area is initialized, setting event trigger thresholds and data processing flows. For example, when the event trigger probability output by the first composite layer exceeds 80%, the inverse reasoning and reasoning operations of the second composite layer are automatically activated. In addition, it is necessary to establish a mapping relationship between the computing area and the storage area, connect the historical causal chain model of the library storage sub-area and the current detection data of the real-time storage sub-area to the computing link, activate the memory-computing integration function of the memristor by initializing the instruction set, so that the computing area can call the stored data in real time and perform parallel inference operations, and finally form a composite computing architecture with self-driven detection capability.

[0055] In one possible implementation, step S200 further includes:

[0056] The composite memristor includes a computing area and a storage area, wherein the storage area includes a library storage sub-area and a real-time storage sub-area; the distributed array arrangement partitioning of the storage area is executed at the node level to initialize the storage area; and a cross-cascade mapping between the storage area and the computing area is established as the composite memristor.

[0057] Specifically, the composite memristor includes a computation area and a storage area, with the storage area further divided into a library storage sub-area and a real-time storage sub-area. The library storage sub-area stores static information such as historically accumulated defect event data, event causal chain models, and prior feature knowledge bases, providing knowledge support for defect detection reasoning. The real-time storage sub-area is responsible for caching dynamic data such as real-time collected detection data, intermediate calculation results, and triggered event information during production line operation, ensuring real-time data interaction and processing during the detection process. This storage area division design enables the composite memristor to efficiently manage different types of data, providing accurate and timely data support for defect composite logic detection in the computation area.

[0058] When initializing the distributed array layout of the storage area at the node level, the storage area is first divided into a library storage sub-area and a real-time storage sub-area. Then, using each computing node as the basic unit, a distributed array layout is performed on these two sub-areas. For the library storage sub-area, partitions are created according to different defect event types, causal chain models, and other knowledge categories. Each partition corresponds to a specific group of nodes for storing historically accumulated static data. The real-time storage sub-area is divided into multiple real-time cache partitions based on the distribution of production line inspection nodes and data acquisition frequency. Each partition is associated with a corresponding computing node, achieving efficient caching of real-time inspection data. During initialization, a unique address identifier is assigned to each storage partition, establishing a mapping relationship between nodes and storage partitions. Simultaneously, parameters such as read / write permissions for the storage medium, data storage format, and cache update strategy are configured to ensure efficient data interaction between the storage area and the computing area, laying the foundation for the in-memory computing functionality of the composite memristor.

[0059] When establishing a cross-cascaded mapping between the storage and computing areas to form a composite memristor, bidirectional data paths are first established between the library storage sub-area and the real-time storage sub-area of ​​the storage area and the first and second composite layers of the computing area, respectively. Data such as historical causal chain models and prior feature knowledge bases stored in the library storage sub-area are associated with the second and third computing nodes in the computing area, responsible for event backward inference and forward deduction, through a mapping mechanism, enabling the computing nodes to call static knowledge for reasoning in real time. The real-time storage sub-area establishes a dynamic mapping with the prior feature detection nodes of the first composite layer and the event reasoning nodes of the second composite layer, ensuring that real-time detection data can synchronously flow into the computing nodes for computation. Simultaneously, by configuring the cross-cascaded control logic and setting data interaction rules, such as automatically activating the mapping link between the real-time storage sub-area and the second composite layer when the first composite layer triggers event detection, the prior data is transmitted to the event reasoning node, forming a closed-loop "storage-computing" interaction system. This ultimately constitutes a composite memristor with integrated storage and computing capabilities, enabling efficient detection and reasoning of production line defects.

[0060] In one possible implementation, step S300 further includes:

[0061] Step S310: As the production line runs, the first calculation node of the first composite layer is triggered to detect and determine the prior data.

[0062] Step S320: The prior data is transmitted back and stored in the real-time storage sub-area as streaming data. Event trigger probability determination based on the first computing node is executed synchronously to determine the prior result.

[0063] Step S330: If the prior result is not triggered, terminate the detection thread.

[0064] Specifically, as the production line operates, the first computing node of the first composite layer is triggered in real time. Each node, based on corresponding prior features (such as equipment vibration frequency, temperature threshold, etc.), collects production line operation data in real time through a precursor detection array. After signal filtering and feature extraction, the data is determined as prior data. During this process, the computing nodes continuously monitor key production line indicators and compare them with a preset prior feature library to ensure the effectiveness and relevance of the collected data, providing accurate data support for subsequent event trigger probability determination.

[0065] Prior data is transmitted back to the real-time storage sub-area of ​​the composite memristor via a high-speed data interface. Streaming data processing technology is employed, with time-series data indexed for block caching, ensuring real-time data writing while supporting subsequent random access. During storage, the first computing node synchronously invokes the built-in Bayesian probability model to compare the real-time transmitted prior feature data (such as vibration frequency and temperature values) with historical defect event samples in the library storage sub-area, calculating the posterior probability of the current prior feature combination triggering a defect event. By setting a threshold (e.g., a probability value ≥ 0.7 is considered a trigger), the prior result is ultimately determined and output to the detection logic control unit, providing a decision-making basis for whether to activate the second composite layer.

[0066] If the prior result determined based on the event trigger probability of the first computing node is that no event has been triggered, the detection thread will be automatically terminated. When the first computing node analyzes the prior data returned to the real-time storage sub-area and determines that the probability of the current prior feature combination triggering a defect event is lower than a preset threshold, it will send a termination command to the detection logic control unit. This unit will then cut off the data transmission link between the first composite layer and the second composite layer, stop subsequent events such as reverse inference, targeted detection, and event deduction, and release the relevant computing nodes and storage resources, causing the detection system to enter a low-power standby state until the next detection cycle or a new trigger condition occurs, at which point the detection process will be restarted.

[0067] In one possible implementation, step S330 further includes:

[0068] Step S331: If the prior result is a trigger, based on the second computing node of the second composite layer, reverse the event characteristics of the prior event, drive the detection device to perform event-oriented detection, and determine the event data.

[0069] Step S332: The event data is transmitted back and stored in the real-time storage sub-area as a stream. Event inference based on the third computing node is executed synchronously to determine the event detection result.

[0070] Specifically, if the prior result is a trigger, based on the second computing node of the second composite layer, reverse causal chain reasoning is performed on the prior event. That is, starting from the triggered prior event, the underlying factors that led to the event are traced back to deduce the corresponding event characteristics, such as abnormal equipment operating parameters or deviations in process indicators. Subsequently, targeted detection instructions are generated based on the deduced event characteristics to drive the detection equipment in the production line (such as sensors, image recognition devices, etc.) to perform targeted detection on specific locations or parameters, collect detailed data related to the event in real time, and after signal processing and feature extraction, it is identified as event data, providing accurate raw data support for subsequent event deduction.

[0071] Event data is transmitted back to the real-time storage sub-area of ​​the composite memristor via a high-speed serial interface. Real-time sharded storage is implemented using the Apache Kafka stream processing framework, with indexes built based on both timestamps and event types to ensure continuous streaming while supporting millisecond-level retrieval. During storage, the third compute node synchronously loads the event causal chain model from the library storage sub-area and uses a graph neural network algorithm to perform forward inference on the event data. For example, if abnormal bearing temperature is detected, the probability of occurrence and impact range of each level of event are calculated by traversing the causal chain path of "temperature increase → lubrication failure → component wear." During the inference process, historical similar event data from the real-time storage sub-area is dynamically invoked as a reference. A probability distribution model of event development is generated through Monte Carlo simulation, ultimately outputting event detection results including event type, severity, and evolution trend, providing quantitative basis for defect alarms.

[0072] In one possible implementation, step S300 further includes:

[0073] Step S340: Identify the event detection results. If a fault event exists, generate alarm information based on the fault status.

[0074] Step S350: Locate the event detection array based on the event detection results, execute the front-end alarm response based on the alarm information, and simultaneously execute the interface pop-up alarm based on the mobile terminal.

[0075] Specifically, a fault recognition neural network model deployed in the composite memristor computing area performs real-time feature extraction and classification of event detection results. This model, trained based on historical fault data, can identify various fault modes such as abnormal vibration spectra and abrupt temperature curve changes. When an event meeting a preset fault threshold is detected (e.g., bearing temperature exceeding 120°C for 5 minutes), the complete causal chain data of the event is automatically retrieved from the real-time storage sub-area, including the fault occurrence timestamp, relevant detection node numbers, and parameter fluctuation curves. Differentiated alarm information is generated based on the severity level of the fault (mild, medium, severe). For example, a severe fault message includes the text description "The motor at workstation A3 on the production line is overheating; immediate shutdown and maintenance are recommended," along with a 3D coordinate map of the fault location and links to historical similar fault handling solutions. The alarm information format follows the ISO 15218 industrial alarm standard to ensure consistent resolution across different terminal devices.

[0076] Based on the event location coordinates and detection node numbers recorded in the event detection results, the physical location of the corresponding event detection array is quickly located using a pre-established detection array topology mapping table. Once the location is determined, a command is immediately sent to the front-end alarm devices in the area where the array is located, triggering audible and visual alarm devices (such as flashing red warning lights and a high-frequency alarm sound from a buzzer). Simultaneously, the alarm information is pushed to the central control system via industrial Ethernet, displaying a real-time view of the fault area on the production line monitoring interface. At the same time, an alarm push is sent to the bound mobile terminal (such as a smartphone or tablet) using the MQTT protocol. Upon receiving the information, the mobile terminal automatically pops up a pop-up interface containing fault details (such as fault type, impact range, and handling suggestions). The interface also includes shortcut buttons for one-click call to the maintenance team and viewing historical fault solutions, ensuring that relevant personnel can promptly obtain alarm information and take appropriate measures regardless of their location.

[0077] Example 2 is based on the same inventive concept as the manufacturing line defect detection method based on big data intelligent algorithms in the previous examples, such as... Figure 2 As shown, this application provides a manufacturing line defect detection system based on big data intelligent algorithms. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0078] The detection cycle chain construction module 10 is used to acquire project production line information, locate prior features-defect events through knowledge reasoning, and construct the detection cycle chain.

[0079] The production line detection array deployment module 20 is used to deploy the production line detection array according to the detection cycle chain, deploy the first computing node based on event triggering based on prior features, deploy the second computing node based on directional detection inference based on event black box, deploy the third computing node based on causal inference based on directional detection, and initialize the memristor with the cross array of the nodes as a composite memristor, wherein the initialization standard includes a storage area and a computing area.

[0080] The event defect alarm module 30 is used to establish array communication interaction between the production line detection array and the composite memristor, and to perform distributed self-driven defect composite logic detection management and execute event defect alarms as the project production progresses.

[0081] Furthermore, the system is also used to implement the following functions:

[0082] According to the detection cycle chain, a first detection node is identified, wherein the first detection node is any node in the detection cycle chain; for the first detection node, a first prior feature and a first defect event are determined; the detection type and production line detection location based on the first prior feature are determined and set as a type I code; the detection type and production line detection location based on the first defect event are determined and set as a type II code; according to the type I code and the type II code, the front-end deployment of the first production line detection array is executed; wherein the corresponding prior feature and the defect event satisfy a preset occurrence interval, and the production line detection array includes a precursor detection array and an event detection array.

[0083] Furthermore, the system is also used to implement the following functions:

[0084] The first computing node is deployed based on prior feature verification and event probability reasoning; for defective events, event causal chains are mined, wherein there is a one-to-one correspondence between defective events and event causal chains; and the second and third computing nodes are deployed based on event reasoning based on the event causal chains.

[0085] Furthermore, the system is also used to implement the following functions:

[0086] The event causal chain is reversed to deduce detection features from defect events, and a second computing node is deployed; the event causal chain is forwarded to perform event deduction based on detection features, and a third computing node is deployed.

[0087] Furthermore, the system is also used to implement the following functions:

[0088] The first computing nodes are arrayed to determine the first composite layer, wherein each first computing node corresponds to a prior feature; the second and third computing nodes are arrayed and cascaded to determine the second composite layer, wherein the first composite layer triggers the second composite layer; the first and second composite layers are cascaded to initialize the computing region of the memristor.

[0089] Furthermore, the system is also used to implement the following functions:

[0090] The composite memristor includes a computing area and a storage area, wherein the storage area includes a library storage sub-area and a real-time storage sub-area; the distributed array arrangement partitioning of the storage area is executed at the node level to initialize the storage area; and a cross-cascade mapping between the storage area and the computing area is established as the composite memristor.

[0091] Furthermore, the system is also used to implement the following functions:

[0092] As the production line operates, the first computing node of the first composite layer is triggered to detect and determine prior data; the prior data is transmitted back and stored in the real-time storage sub-area for streaming data storage, and the event trigger probability determination based on the first computing node is executed synchronously to determine the prior result; if the prior result is that it has not been triggered, the detection thread is terminated.

[0093] Furthermore, the system is also used to implement the following functions:

[0094] If the prior result is triggered, the event characteristics are inversely deduced from the prior event based on the second computing node of the second composite layer, and the detection device is driven to perform event-oriented detection to determine the event data; the event data is transmitted back and stored in the real-time storage sub-area for streaming data storage, and event deduction based on the third computing node is executed synchronously to determine the event detection result.

[0095] Furthermore, the system is also used to implement the following functions:

[0096] The event detection results are identified. If a fault event exists, alarm information is generated based on the fault status. The event detection array based on the event detection results is located, and a front-end alarm response based on the alarm information is executed. Simultaneously, an interface pop-up alarm based on the mobile terminal is executed.

[0097] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0098] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0099] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for manufacturing line defect detection based on big data intelligent algorithm, characterized in that, The method comprises: acquiring project production line information, positioning prior feature-defect events by performing knowledge reasoning, and constructing a detection cycle chain; deploying a production line detection array according to the detection cycle chain, deploying a first computing node based on prior feature event triggering, deploying a second computing node based on event black box directional detection reasoning, and deploying a third computing node based on directional detection causal reasoning, initializing a memistor in a cross array of nodes as a composite memistor, wherein the initialization standard comprises a storage area and a calculation area; establishing array communication interaction between the production line detection array and the composite memistor, performing distributed self-driven defect composite logic detection management, and executing event defect alarm along with project production progress; the production line detection array deployment according to the detection cycle chain comprises: identifying a first detection node according to the detection cycle chain, wherein the first detection node is any detection cycle chain node; determining a first prior feature and a first defect event for the first detection node; determining the detection type and production line detection location based on the first prior feature, and setting it as a first type of code; determining the detection type and production line detection location based on the first defect event, and setting it as a second type of code; executing front-end deployment of a first production line detection array according to the first type of code and the second type of code; wherein the corresponding prior feature and the defect event satisfy a preset occurrence interval, and the production line detection array comprises a precursor detection array and an event detection array.

2. The big data intelligent algorithm based manufacturing production line defect detection method of claim 1, wherein, deploying a first computing node based on prior feature event triggering, deploying a second computing node based on event black box directional detection reasoning, and deploying a third computing node based on directional detection causal reasoning, comprises: deploying a first computing node based on prior feature detection-event probability reasoning; mining an event causal chain for a defect event, wherein the defect event and the event causal chain correspond one-to-one; deploying a second computing node and a third computing node based on event reasoning based on the event causal chain.

3. The big data intelligent algorithm based manufacturing production line defect detection method of claim 2, wherein, deploying a second computing node and a third computing node comprises: deploying a second computing node by inversely pushing a detection feature from a defect event based on the event causal chain; deploying a third computing node by performing event deduction based on a detection feature.

4. The big data intelligent algorithm based manufacturing production line defect detection method of claim 3, wherein, initializing a memistor in a cross array of nodes comprises: array arranging first computing nodes to determine a first composite layer, wherein each first computing node corresponds to a prior feature; array arranging and cascading second computing nodes and third computing nodes to determine a second composite layer, wherein the first composite layer triggers the second composite layer; cascading the first composite layer and the second composite layer to initialize the calculation area of the memistor.

5. The big data intelligent algorithm based manufacturing production line defect detection method of claim 4, wherein, The composite memistor comprises a calculation area and a storage area, wherein the storage area comprises a library storage sub-area and a real-time storage sub-area; performing distributed array arrangement partitioning of the storage area by node order to initialize the storage area; establishing cross-cascading mapping of the storage area and the calculation area as the composite memistor.

6. The big data intelligent algorithm based manufacturing production line defect detection method of claim 1, wherein, performing distributed self-driven defect composite logic detection management along with project production progress comprises: As the production line operates, the first calculation node of the first composite layer is triggered to detect and determine the prior data. The prior data is sent back and stored in the real-time storage sub-area as streaming data. Simultaneously, the event trigger probability determination based on the first computing node is executed to determine the prior result. If the prior result is not triggered, the detection thread is terminated.

7. The big data intelligent algorithm based manufacturing production line defect detection method of claim 6, wherein, If the prior result is a trigger, the event characteristics are inversely deduced from the prior event based on the second computing node of the second composite layer, and the detection device is driven to perform event-oriented detection to determine the event data; The event data is transmitted back and stored in the real-time storage sub-area as streaming data. Event inference based on the third computing node is executed synchronously to determine the event detection result.

8. The big data intelligent algorithm based manufacturing production line defect detection method of claim 7, wherein, Execution event defect alerts include: The event detection results are identified, and if a fault event exists, alarm information is generated based on the fault status. The event detection array is located based on the event detection results, and a front-end alarm response based on the alarm information is executed, while an interface pop-up alarm based on the mobile terminal is executed simultaneously.

9. A manufacturing line defect detection system based on big data intelligent algorithm, characterized in that, The system is used to implement the manufacturing production line defect detection method based on big data intelligent algorithms as described in any one of claims 1-8, and the system comprises: The detection cycle chain construction module is used to acquire project production line information, locate prior features-defect events through knowledge reasoning, and construct the detection cycle chain. The production line detection array deployment module is used to deploy the production line detection array according to the detection cycle chain, deploy the first computing node based on event triggering based on prior features, deploy the second computing node based on directional detection inference based on event black box, deploy the third computing node based on causal inference based on directional detection, and initialize the memristor with the cross array of the nodes as a composite memristor, wherein the initialization standard includes the storage area and the computing area. The event defect alarm module is used to establish array communication interaction between the production line detection array and the composite memristor, and to perform distributed self-driven defect composite logic detection management and execute event defect alarms as the project production progresses.

Images