Fault repair method and system based on big data analysis

Abstract

The application relates to the technical field of data processing, in particular to a fault repair method and system based on big data analysis. The method comprises the following steps: in response to receiving a new repair request, obtaining a state deviation degree based on the difference between real-time operation parameters and historical operation parameters of various types; converting text information in the new repair request into a real-time fault vector, and calculating the semantic similarity between the real-time fault vector and historical text vectors of various fault types; for any fault type, obtaining a comprehensive evaluation index of the fault type by weighting and fusing the state deviation degree and the semantic similarity corresponding to the fault type, determining the fault type with the highest evaluation index as the diagnosis result of the current repair, and intelligently assigning a repair task according to the diagnosis result. The application has the effects of improving the timeliness of maintenance and improving the utilization rate of human resources.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a fault reporting method and system based on big data analysis. Background Technology

[0002] In modern industrial and commercial environments, the stable operation of industrial equipment is crucial. Traditional fault reporting processes rely heavily on manual handling: field personnel manually create work orders after discovering an anomaly, and dispatchers rely on personal experience and the textual descriptions on the work orders to diagnose the fault and assign engineers. This model has many inconveniences. For example, due to the subjectivity and ambiguity of user descriptions, dispatchers are prone to errors in judgment, resulting in dispatched engineers with mismatched skills or without the correct spare parts, increasing the rate of secondary repairs and costs. At the same time, the manual dispatch process is cumbersome, slow in response, and difficult to cope with periods of high failure rates. It also fails to provide a global perspective on resource allocation, often resulting in some engineers being overloaded while others are idle, leading to a waste of human resources.

[0003] To address these issues, the industry has begun to explore the introduction of data analytics. In fault diagnosis, some solutions utilize real-time equipment operating data, employing statistical models such as Mahalanobis distance to determine if the equipment's status deviates from normal limits. Once equipment malfunction is confirmed, intelligent task dispatch is implemented, typically using optimization algorithms such as the Hungarian algorithm to solve the task allocation problem, aiming to find the lowest-cost allocation solution.

[0004] In related technologies, methods for judging equipment malfunctions only focus on numerical changes, leading to false positives and false negatives during the malfunction detection process. Furthermore, the dispatching process only considers distance or the availability of engineers, failing to comprehensively allocate human resources. In summary, the fault reporting methods in these technologies suffer from untimely repairs and low utilization of human resources in actual operation. Summary of the Invention

[0005] To improve the timeliness of equipment maintenance and the utilization rate of human resources, this application provides a fault reporting method and system based on big data analysis.

[0006] Firstly, this application provides a fault reporting method based on big data analysis, employing the following technical solution:

[0007] The fault reporting method based on big data analysis includes: obtaining historical operating parameters and maintenance records for various types of equipment faults; and using a pre-trained natural language processing model to obtain historical text vectors for each fault type based on the text information in the maintenance records.

[0008] In response to a new repair request, the system obtains the state deviation degree based on the difference between real-time operating parameters and historical operating parameters for each type; it converts the text information in the new repair request into a real-time fault vector and calculates the semantic similarity between the real-time fault vector and the historical text vectors for each fault type; for any fault type, it obtains a comprehensive evaluation index for that fault type by weighted fusion of the state deviation degree and semantic similarity, and determines the fault type with the highest evaluation index as the diagnosis result for the current repair request; and it intelligently dispatches repair tasks based on the diagnosis results.

[0009] The steps for intelligently dispatching repair tasks based on diagnostic results include: constructing a dispatch cost function, which includes the estimated time cost for the engineer to reach the task location, the matching cost between the engineer's skills and task requirements, and the engineer's current workload cost; and using a preset allocation algorithm to determine an allocation scheme that minimizes the dispatch cost.

[0010] In this application, the state deviation is calculated using quantitative data of equipment operation, and semantic similarity is calculated using textual information from user repair requests. Then, the state deviation and semantic similarity are weighted and fused to achieve a dual-channel, high-precision comprehensive diagnosis of fault types, significantly reducing the false positive rate. Furthermore, a multi-dimensional dispatch cost function is constructed based on factors such as the engineer's travel time to the task location, the engineer's skills, and the engineer's workload. This ensures that suitable engineers are dispatched for high-quality repairs, improving the utilization rate of human resources and the timeliness of repairs.

[0011] Optionally, the pre-trained natural language processing model is the BERT model.

[0012] By using the BERT model, we can gain a deeper understanding of the semantics of the repair report text and accurately identify synonyms such as "burned" and "eroded", thereby enhancing the overall effect of fault diagnosis.

[0013] Optionally, the dispatch cost function also includes spare parts acquisition cost; the spare parts acquisition cost is calculated based on the engineer's current location, the location of the spare parts warehouse, and real-time navigation information of the task location.

[0014] Taking into account the time and cost for engineers to obtain spare parts makes dispatching decisions more comprehensive and refined.

[0015] Optionally, the default allocation algorithm is the Hungarian algorithm.

[0016] Optionally, it may also include: after dispatching a work order, retrieving standard operating procedures, historical repair cases, or technical drawings related to the diagnostic results from the knowledge base and pushing them to the terminal device of the assigned engineer.

[0017] After dispatching a repair order, a function has been added to push relevant knowledge (such as standard operating procedures and historical cases) to engineers. Compared with engineers relying solely on personal experience for repairs, this provides engineers with immediate and standardized operation guidance and reference, which greatly improves the standardization and efficiency of repair work. In particular, it can help new engineers get started quickly, shorten repair time, and ensure repair quality.

[0018] Optionally, the weight coefficients of each cost item in the dispatch cost function are dynamically adjusted using a reinforcement learning model; the input of the reinforcement learning model is the current status index of the operation and maintenance system, and the output is a set of updated weight coefficients.

[0019] A reinforcement learning model is used to dynamically adjust the weights of various costs in the dispatch cost function, enabling the dispatch strategy to adapt adaptively.

[0020] Optionally, the calculation steps for matching the engineer's skills with the task requirements include: obtaining the skill level requirements for the task and the actual skill level of the engineer, and accumulating the results based on the product of the difference between the two skill levels and a preset skill importance coefficient.

[0021] Optionally, it also includes receiving feedback data submitted by engineers after the repair is completed, which includes the finally confirmed cause of the fault and the actual repair operations.

[0022] Secondly, this application provides a fault reporting system based on big data analysis, employing the following technical solution:

[0023] The fault reporting system based on big data analysis includes a processor and a memory. The memory stores computer program instructions, which are executed by the processor to implement the fault reporting method based on big data analysis as described above.

[0024] The aforementioned fault reporting method based on big data analysis is generated into a computer program and stored in memory so that it can be loaded and executed by the processor. Thus, a system can be created based on the memory and processor for easy use.

[0025] This application has the following technical effects:

[0026] By integrating equipment operation data (status deviation) and repair request text information (semantic similarity) for dual-channel fault diagnosis, accuracy has been significantly improved. Simultaneously, a multi-dimensional dispatch cost function has been constructed, and the globally optimal allocation of human and material resources has been achieved through algorithm optimization. Attached Figure Description

[0027] Figure 1 This is a flowchart of the fault reporting method based on big data analysis in an embodiment of this application. Detailed Implementation

[0028] This application discloses a fault reporting method based on big data analysis. By deeply integrating real-time equipment operating data with historical maintenance knowledge, it constructs an end-to-end intelligent management process, from intelligent fault diagnosis to globally optimal dispatching, and then to a knowledge base self-learning closed loop. Compared to the traditional passive response maintenance model that relies on human experience, this method can significantly improve the accuracy of fault diagnosis, shorten maintenance response time, and achieve dynamic optimal allocation of maintenance human and material resources, thereby reducing the overall maintenance cost of the enterprise and improving the reliability of equipment operation.

[0029] Reference Figure 1 The fault reporting method based on big data analysis includes steps S1-S6.

[0030] S1: Obtain historical operating parameters and maintenance records for various types of equipment faults.

[0031] The device under monitoring is equipped with an IoT gateway, which can collect multi-dimensional operating parameters in real time during device operation. Therefore, it can obtain historical fault data for various types of failures that occurred during the device's history. This historical fault data can include voltage, current, temperature, three-phase imbalance, power factor, harmonic content, and instantaneous surge voltage. The acquisition frequency can be set according to the variation characteristics of the parameters; for example, for rapidly changing parameters such as voltage and current, a higher acquisition frequency is preferred. times / second; for slowly changing parameters such as temperature, the sampling frequency can be set to... times per minute.

[0032] To eliminate the differences caused by different data sources and formats, the collected raw data needs to be standardized.

[0033] For numerical equipment operating parameters, the max-min normalization method is used to map all parameter values ​​to... Within the specified interval, the normalized historical fault data is used as historical operating parameters. This process eliminates the differences between different physical dimensions, ensuring that each parameter has equal initial weight in subsequent model calculations.

[0034] Finally, by using methods such as linear interpolation or time window aggregation, data points collected at different frequencies are aligned to a unified timestamp to form a wide-table format device status snapshot dataset.

[0035] Extract historical maintenance records from the enterprise's internal operations and maintenance management system (such as CMMS or EAM system).

[0036] S2: Based on the text information in the maintenance records, a pre-trained natural language processing model is used to obtain the historical text vectors of each fault type.

[0037] Next, text information is extracted from the maintenance records. For example, structured fields such as maintenance date and replacement part model are extracted. At the same time, unstructured text data such as maintenance records and descriptions of fault symptoms are deeply mined.

[0038] For unstructured text data, this embodiment uses a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model for processing. Specifically, fault feature words (such as burnout, sparking) extracted from maintenance records or complete fault description statements can be input into the BERT model, converting them into high-dimensional semantic vectors. For example, using the BERT-base-chinese model, each text input can be converted into a... A 3D vector. This processing method enables machines to understand the high semantic similarity between "burnout" and "ablation," overcoming the limitations of traditional keyword matching methods. As a preferred approach, each known fault type in history is represented... All relevant maintenance record texts are aggregated, processed by the BERT model, and their average semantic vectors are calculated to obtain historical text vectors representing this type of fault, which are then stored in the knowledge base.

[0039] S3: In response to receiving a new repair request, obtain the status deviation based on the difference between real-time operating parameters and historical operating parameters of various types.

[0040] A fault that is about to occur or has already occurred will inevitably leave traces at the physical level (abnormal operating parameters). Therefore, in this embodiment, after receiving a repair request, the current real-time operating parameters of the equipment are first obtained and compared with the historical operating parameters corresponding to various fault types to reflect the type of fault that is currently occurring in the equipment.

[0041] The steps for obtaining the state deviation based on the difference between real-time operating parameters and various types of historical operating parameters include: both real-time operating parameters and historical operating parameters include multi-dimensional data, wherein the multi-dimensional data in the real-time operating parameters constitute the real-time feature vector, the mean of each dimension of the historical operating parameters is used as the historical fault vector, and the Mahalanobis distance between the real-time feature vector and the historical fault vector is used as the state deviation.

[0042] Mahalanobis distance can take into account the correlation between parameters and is sensitive to the shape of the data distribution. Therefore, in this embodiment, the Mahalanobis distance between the real-time feature vector and the historical fault vector is used as the state deviation. This index reflects the difference between the current state of the device and the fault type. The larger the difference (state deviation), the less the current state of the device conforms to the fault type.

[0043] S4: Convert the text information in the new repair request into a real-time fault vector, and calculate the semantic similarity between the real-time fault vector and the historical text vector of each fault type.

[0044] After a fault occurs, similarities exist not only in equipment operating parameters but also in the experience of the staff. Therefore, upon receiving a new repair request, the fault description text is converted into a real-time fault vector using the NLP model (such as BERT) used in step S1. Step S2, which converts the fault into a real-time fault vector, is similar and will not be elaborated further. For any fault type, the cosine similarity between the historical fault vector and the real-time fault vector is used as the semantic similarity.

[0045] S5: For any fault type, the comprehensive evaluation index of the fault type is obtained by weighted fusion of the state deviation degree and semantic similarity corresponding to the fault type, and the fault type with the highest evaluation index is determined as the diagnosis result of the current repair.

[0046] State deviation and semantic similarity are two indicators for judging whether the current device state belongs to the fault. Therefore, the two are fused to comprehensively judge the actual operating state of the device.

[0047] The formula for calculating the comprehensive health assessment index can be expressed as:

[0048] In the formula, Indicates a point in time Equipment malfunction The comprehensive evaluation index. The higher the value, the greater the probability of the failure occurring; Indicates time Real-time feature vectors and fault types The state deviation between historical fault vectors is considered here because the probability of this value being 0 is extremely small. However, in other embodiments, a hyperparameter can be set, the sum of the hyperparameter and the state deviation can be calculated, and then the sum of the hyperparameter and the state deviation can be raised to the power of -1. In the formula This is a hyperparameter. The larger the value, the faster the time... The less the fault occurs, the less it is considered a fault. ; Indicates time Real-time fault vectors and fault types Semantic similarity between historical text vectors; This represents the Sigmoid function, used to normalize state deviation. This represents the weighting coefficient, used to balance the importance of real-time operational data and historical text information in fault diagnosis. It is automatically fine-tuned using gradient descent. The value of allows the adjusted model to improve its assessment score of the real cause of the failure in the next round of calculations. This value can also be set by those skilled in the art based on experience.

[0049] Will Mapped to via the Sigmoid activation function Intervals, making them similar to Being on the same scale makes weighting easier. It is a weighting coefficient used to balance the importance of real-time operational data and historical text information in fault diagnosis. It is automatically fine-tuned using gradient descent. The value of allows the adjusted model to improve the accuracy of analyzing the real causes of failures in the next round of calculations.

[0050] Calculate the comprehensive evaluation index for all known fault types, and select the one with the highest comprehensive evaluation index as the diagnostic result.

[0051] S6: Intelligent dispatch of repair tasks based on diagnostic results.

[0052] During the maintenance process, the availability of spare parts in the spare parts warehouse, the distance between the engineer and the equipment location, and the engineer's skills all affect the timeliness of maintenance. In order to comprehensively consider the influence of multiple factors, this embodiment constructs a dispatch cost function to represent the dispatch cost of assigning a task to a maintenance engineer.

[0053] Specifically, the dispatch cost function includes the estimated time cost for the engineer to reach the task location, the matching cost between the engineer's skills and task requirements, and the engineer's current workload cost; a preset allocation algorithm is used to determine an allocation scheme that minimizes the dispatch cost.

[0054] The formula (cost function) for calculating the dispatch cost of assigning any task to any engineer can be expressed as:

[0055] ;in, Indicates task Dispatch the order to the engineer The cost of dispatching orders; As the first weighting coefficient, The second weighting coefficients are summed as follows: ; This is the third weighting coefficient; It is the fourth weighting coefficient.

[0056] The estimated time cost for engineers to reach the task location is obtained by calling the real-time map navigation service API. The estimated time cost for all available engineers is then normalized using a maximum-min method to obtain the result. The fastest arriving engineer scored as follows: The slowest score is .

[0057] This represents the matching cost between an engineer's skills and the requirements of a task; it is mainly calculated by the degree of matching between the skills required for the task and the skills possessed by the engineer. The calculation formula can be expressed as: ; For matching costs; Indicates task The collection of all the necessary skills; Indicates the skills required for the task. The level; This indicates the level of skill 's' possessed by the engineer; Skill The importance of the task is represented by a value between 0 and 1. If an engineer simply lacks a skill with a low importance coefficient, then... The impact on cost is relatively small; given suitable distance and load conditions, the system may still assign him to perform tasks.

[0058] This represents the current workload cost of an engineer; it primarily reflects whether an engineer is available. In this embodiment, for any engineer, the workload cost is represented by the estimated total completion time of their pending tasks. For example, if an engineer currently has pending tasks A and B, and based on historical experience, task A takes 1 hour to complete and task B takes 2 hours to complete, then the engineer's estimated total completion time is 3 hours. The estimated total completion time for all engineers is calculated and normalized to obtain the workload cost for each engineer.

[0059] This represents the cost of acquiring spare parts, calculated based on real-time navigation information of the engineer's current location, the location of the spare parts depot, and the task location. In this embodiment, the time taken for the engineer to travel to the spare parts depot and then to the task location is calculated, and the time is used to reflect the cost of acquiring spare parts. Specifically, the formula for calculating the cost of acquiring spare parts can be expressed as:

[0060] In the formula, Indicates the cost of acquiring spare parts; Indicates engineer Location; This indicates the time required for the engineer to reach the spare parts warehouse; Indicate the mission location; Indicates from spare parts warehouse Time to reach the mission location. This part is calculated when spare parts are needed for repairs; otherwise, it is not calculated. Defined as 0.

[0061] Weighting coefficient The system dynamically adjusts its settings through reinforcement learning. It takes the current operational status (such as the number of pending work orders and average repair time) as input and outputs a set of optimal weights. For example, during periods of high failure incidence, the system might increase... The weighting is prioritized to ensure response speed; when handling complex faults, it may improve... The weighting is used to ensure a one-time repair rate. In other embodiments... It can also be set by those skilled in the art based on experience.

[0062] In one environment, based on a maintenance statistics table, when there are One pending task and When there are available engineers, the system calculates one. The cost matrix is ​​constructed, where each element represents the dispatch cost between each engineer and each task. Then, the Hungarian Algorithm is used to solve the assignment problem, finding the dispatch scheme that minimizes the overall total cost.

[0063] After generating a work order, the system automatically adds auxiliary information related to the type of fault to the electronic work order, including standard operating procedures (SOPs), historical successful repair cases, technical drawings, and a detailed list of required tools and spare parts, which greatly improves the work efficiency and standardization of engineers, especially new engineers.

[0064] After completing the repairs on-site, engineers confirm and provide feedback on the results via a mobile app. Feedback data includes the confirmed root cause of the fault, the actual repair steps performed, the actual time taken, and the actual spare parts replaced.

[0065] In this step, high-quality data verified by front-line engineers is fed back into the database as new training samples for continuous optimization of weight coefficient values ​​in step S5. It can also be used to periodically update historical operating parameters, continuously improving the accuracy of equipment fault diagnosis and the timeliness of equipment maintenance.

[0066] This application also discloses a fault reporting system based on big data analysis, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the fault reporting method based on big data analysis according to this application.

[0067] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0068] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A fault reporting method based on big data analysis, characterized in that, include: Obtain historical operating parameters and maintenance records for various types of equipment failures; based on the text information in the maintenance records, use a pre-trained natural language processing model to obtain historical text vectors for each failure type; In response to a new repair request, the status deviation is obtained based on the difference between real-time operating parameters and historical operating parameters of each type; the text information in the new repair request is converted into a real-time fault vector, and the semantic similarity between the real-time fault vector and the historical text vector of each fault type is calculated; for any fault type, the weighted fusion of the status deviation and semantic similarity corresponding to the fault type is used to obtain the comprehensive evaluation index of the fault type, and the fault type with the highest evaluation index is determined as the diagnosis result of the current repair. Intelligent dispatching of repair tasks based on diagnostic results; The steps for intelligently dispatching repair tasks based on diagnostic results include: constructing a dispatch cost function, which includes the estimated time cost for the engineer to reach the task location, the matching cost between the engineer's skills and task requirements, and the engineer's current workload cost; and using a preset allocation algorithm to determine an allocation scheme that minimizes the dispatch cost. The steps for obtaining the state deviation based on the difference between real-time operating parameters and various types of historical operating parameters include: both real-time operating parameters and historical operating parameters include multi-dimensional data, wherein the multi-dimensional data in the real-time operating parameters constitute a real-time feature vector, the mean of each dimension of the historical operating parameters is used as the historical fault vector, and the Mahalanobis distance between the real-time feature vector and the historical fault vector is used as the state deviation. The default allocation algorithm is the Hungarian algorithm; The steps for calculating the matching cost between engineer skills and task requirements include: obtaining the skill level requirements for the task and the actual skill level of the engineer, and then summing them up based on the product of the difference between the two skill levels and a preset skill importance coefficient. The formula for calculating the dispatch cost of assigning any task to any engineer can be expressed as: ; Indicates task Dispatch the order to the engineer The cost of dispatching orders; As the first weighting coefficient, This is the second weighting coefficient; This is the third weighting coefficient; This is the fourth weighting coefficient; The estimated time cost for engineers to reach the task location; This represents the cost of matching an engineer's skills with the requirements of the task. This indicates the current workload cost for engineers; This indicates the cost of acquiring spare parts.

2. The fault reporting method based on big data analysis according to claim 1, characterized in that, The pre-trained natural language processing model is the BERT model.

3. The fault reporting method based on big data analysis according to claim 1, characterized in that, The dispatch cost function also includes spare parts acquisition cost; the spare parts acquisition cost is calculated based on the engineer's current location, the location of the spare parts warehouse, and real-time navigation information of the task location.

4. The fault reporting method based on big data analysis according to claim 1, characterized in that, Also includes: After dispatching the order, the system retrieves standard operating procedures, historical repair cases, or technical drawings related to the diagnostic results from the knowledge base and pushes them to the terminal device of the assigned engineer.

5. The fault reporting method based on big data analysis according to claim 1, characterized in that, The weight coefficients of each cost item in the dispatch cost function are dynamically adjusted using a reinforcement learning model; the input of the reinforcement learning model is the current status index of the operation and maintenance system, and the output is a set of updated weight coefficients.

6. The fault reporting method based on big data analysis according to claim 1, characterized in that, Also includes: Receive feedback data submitted by engineers after completing repairs. The feedback data includes the final confirmed cause of the fault and the actual repair operations.

7. A fault reporting system based on big data analysis, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the fault reporting method based on big data analysis according to any one of claims 1-6 is implemented.

Images