Servo controller fault data analysis method and environmental sensing analysis system

Abstract

The present invention discloses a servo controller fault data analysis method and an environmental sensor analysis system, and belongs to the field of data analysis technology. The present invention obtains a variety of logs and comprehensively integrates historical data to provide a basis for analysis. The environmental factor impact evaluation obtained by correlation analysis can accurately quantify the impact of environmental factors on servo controller failures. A fault tracing point of view thermal network is constructed to intuitively present the fault tracing point of view and its correlation strength. The first traceability discrimination thermal matrix obtained provides a quantitative basis for fault tracing. The optimized target operation and maintenance fault tracing model significantly improves the accuracy of fault tracing, and can effectively generate fault tracing points of view for new monitoring information. In this way, the problems of the lack of environmental factor system analysis and the insufficient accuracy of the fault tracing model in the traditional method are overcome, the downtime is reduced, the reliability and maintenance efficiency of the servo control system are improved, and the intelligent operation and maintenance in the field of industrial automation are strongly supported.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a servo controller fault data analysis method and an environmental sensing analysis system. Background Art

[0002] In the field of industrial automation, the stable operation of servo controllers is crucial. However, traditional fault tracing methods have many defects. On the one hand, there is a lack of comprehensive and systematic analysis of environmental factors, and the comprehensive impact of various environmental factors such as temperature, humidity, vibration, etc. on servo controller failures in a closed climate operating environment is often ignored. On the other hand, the existing fault tracing model is not accurate enough. Since it does not fully consider the fault category characteristics in historical data and the complex relationship between environmental factors and faults, it is difficult to accurately trace the cause of the fault when facing actual faults, resulting in low maintenance efficiency, increasing the downtime of the servo control system, and affecting the efficiency and stability of the entire industrial automation production. Summary of the invention

[0003] The present invention provides a servo controller fault data analysis method and an environmental sensing analysis system, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The present invention provides a servo controller fault data analysis method, which is applied to an environmental sensor analysis system. The method comprises: the servo controller fault data analysis method is applied to an environmental sensor analysis system, and the servo controller fault data analysis method comprises: obtaining a priori environmental monitoring log and a past environmental monitoring log, wherein the priori environmental monitoring log comprises at least one priori operating environment sensor monitoring information and a servo controller fault event category feature of the priori operating environment sensor monitoring information, and the past environmental monitoring log comprises a plurality of past operating environment sensor monitoring information; according to the servo controller fault event category feature, a fault performance anomaly involving vector of the priori operating environment sensor monitoring information and a closed climate operation of the past operating environment sensor monitoring information are analyzed; The operation environment description semantics are analyzed for association to obtain the environmental factor impact evaluation of the past operation environment sensor monitoring information; fault tracing is performed based on the environmental factor impact evaluation to obtain the fault tracing viewpoint thermal network of the servo controller fault event in the past operation environment sensor monitoring information; based on the fault tracing viewpoint thermal network, a first tracing discrimination thermal matrix of the fault event triggering feature set in the past operation environment sensor monitoring information is obtained; based on the first tracing discrimination thermal matrix and the past operation environment sensor monitoring information, an initial operation and maintenance fault tracing model is optimized to obtain a target operation and maintenance fault tracing model, wherein the target operation and maintenance fault tracing model is used to generate a fault tracing viewpoint of the servo controller fault event for the operation environment sensor monitoring information to be analyzed.

[0005] The present invention provides an environmental sensing analysis system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0006] The present invention provides a readable storage medium, on which a program or instruction is stored, and the program or instruction implements the steps of the above method when executed by a processor.

[0007] The present invention obtains a variety of logs and comprehensively integrates historical data to provide a basis for analysis. The environmental factor impact evaluation obtained by correlation analysis can accurately quantify the impact of environmental factors on servo controller failures. A fault tracing viewpoint thermal network is constructed to intuitively present the fault tracing viewpoints and their correlation strengths. The first traceability discrimination thermal matrix obtained provides a quantitative basis for fault tracing. The optimized target operation and maintenance fault tracing model significantly improves the accuracy of fault tracing and can effectively generate fault tracing viewpoints for new monitoring information. In this way, the problems of the traditional method's lack of systematic analysis of environmental factors and the insufficient accuracy of the fault tracing model are overcome, downtime is reduced, the reliability and maintenance efficiency of the servo control system are improved, and intelligent operation and maintenance in the field of industrial automation are strongly supported. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 The present invention provides a flow chart of a servo controller fault data analysis method.

[0009] Figure 2 This is a structural schematic diagram of an environmental sensing and analysis system provided by the present invention. DETAILED DESCRIPTION

[0010] Figure 1 A servo controller fault data analysis method is shown, which is applied to an environmental sensing analysis system. The method includes the following steps 120 to 180.

[0011] Step 120: The environmental sensing analysis system obtains prior environmental monitoring logs and past environmental monitoring logs.

[0012] Among them, the prior environment monitoring log includes at least one prior operating environment sensor monitoring information and a servo controller fault event category feature of the prior operating environment sensor monitoring information, and the past environment monitoring log includes multiple past operating environment sensor monitoring information.

[0013] In detail, the task of the environmental sensor analysis system in step 120 is to obtain two types of important log information. The prior environmental monitoring log contains at least one prior operating environment sensor monitoring information and the corresponding servo controller fault event category feature. For example, the prior operating environment sensor monitoring information includes monitoring data of physical quantities such as temperature value (such as 25-30°C is the normal range), humidity value (such as 40%-60% is the appropriate humidity range), and vibration amplitude (such as less than 0.1mm / s is the normal vibration range). At the same time, the servo controller fault event category feature can be a clear fault type classification such as "overheating fault", "short circuit fault caused by excessive humidity", and "loose connection fault caused by vibration".

[0014] The past environment monitoring log contains multiple past operating environment sensor monitoring information. These past operating environment sensor monitoring information are also monitoring records of various parameters in the environment, such as the past measurement values ​​of temperature, humidity, vibration and other parameters at different time points. For example, within a certain period of time, the temperature value is recorded every 10 minutes to form a time series of temperature monitoring data, and the same applies to humidity and vibration data. These data constitute the main content of the past environment monitoring log.

[0015] Step 140: The environmental sensor analysis system performs correlation analysis on the fault performance variation involvement vector of the prior operating environment sensor monitoring information and the closed climate operating environment description semantics of the past operating environment sensor monitoring information based on the servo controller fault event category characteristics, and obtains the environmental factor impact evaluation of the past operating environment sensor monitoring information.

[0016] Furthermore, the task of step 140 is to obtain the environmental factor impact evaluation of the past operating environment sensor monitoring information through correlation analysis. Among them, for the prior operating environment sensor monitoring information, the fault performance variation involvement vector can be understood as a quantitative representation of the degree and direction of the environmental parameters deviating from the normal range under a specific fault event. For example, in the prior operating environment sensor monitoring information of "overheating fault", if the normal temperature range is 25-30°C, and the temperature reaches 40°C when the fault occurs, then in the temperature dimension, the fault performance variation involvement vector may be expressed as a deviation of 10°C in the high temperature direction (for example, expressed as an exemplary difference). The closed climate operating environment description semantics is a semantic description of the environmental state reflected by the past operating environment sensor monitoring information. For example, semantic descriptions such as high temperature, moderate humidity, and low vibration.

[0017] Next, the environmental sensor analysis system conducts correlation analysis on the fault performance variation vector of the prior operating environment sensor monitoring information and the closed climate operating environment description semantics of the past operating environment sensor monitoring information based on the servo controller fault event category characteristics. This process may involve complex algorithm models, such as a probabilistic reasoning model based on a Bayesian network.

[0018] For example, in the Bayesian network, there are nodes representing fault event categories (such as overheating faults, humidity faults, vibration faults, etc.), and there are nodes representing environmental parameters (temperature, humidity, vibration, etc.). The fault performance variation involvement vector in the prior operating environment sensor monitoring information is input into the Bayesian network as known evidence, and the environmental parameter values ​​corresponding to the closed climate operating environment description semantics of the past operating environment sensor monitoring information are also used as variables in the network. Through the Bayesian theorem, the association probability between the past operating environment sensor monitoring information and different fault events is calculated under given prior evidence.

[0019] For example, for a past operating environment sensor monitoring information, the temperature is 35°C (semantically high temperature), the humidity is 50% (semantically moderate humidity), and the vibration is 0.05mm / s (semantically small vibration). Through Bayesian network calculation, the association probability with "overheating fault" is 0.3 (for example, numerical value), the association probability with "humidity fault" is 0.1, and the association probability with "vibration fault" is 0.05, etc. Based on these probability values, the environmental factor impact evaluation of the past operating environment sensor monitoring information is obtained comprehensively. This evaluation can be a comprehensive score or a ranking of the degree of association of different fault types, such as the environmental factor impact evaluation of the information is: "overheating fault" is more likely, "humidity fault" is less likely, and "vibration fault" is extremely unlikely.

[0020] Step 160: The environmental sensor analysis system performs fault tracing based on the environmental factor impact evaluation to obtain a fault tracing viewpoint thermal network of the servo controller fault event in the past operating environment sensor monitoring information; based on the fault tracing viewpoint thermal network, obtain a first tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information.

[0021] In the present invention, step 160 focuses on fault tracing to obtain a fault tracing point of view thermal network and a first tracing discrimination thermal matrix. The fault tracing process is as follows: the environmental sensing analysis system performs fault tracing based on the environmental factor impact evaluation obtained in step 140. Fault tracing is a process of inferring the cause from the result (the possibility of fault reflected by the environmental factor impact evaluation). For example, if the environmental factor impact evaluation shows that the possibility of "overheating fault" is high, then the fault tracing will focus on factors related to temperature, such as possible causes such as heat dissipation system failure, excessive ambient temperature leading to poor heat dissipation, etc.

[0022] In this process, a thermal network of fault tracing viewpoints of servo controller fault events in past operating environment sensor monitoring information is obtained. This thermal network can be understood as a network structure with fault tracing viewpoints as nodes and the strength of association between viewpoints as edge weights. For example, the tracing viewpoints for "overheating fault" may include "heat sink blockage", "fan failure", "ambient temperature is too high", etc. The strength of association between these viewpoints may be quantified based on actual data and analysis results. For example, the strength of association between "heat sink blockage" and "fan failure" is 0.4 (indicating the possibility that both cause overheating faults at the same time), and the strength of association between "heat sink blockage" and "ambient temperature is too high" is 0.2, etc.

[0023] Then, the process of obtaining the first traceability discrimination thermal matrix is ​​as follows: according to the fault traceability viewpoint thermal network, obtain the first traceability discrimination thermal matrix of the fault event trigger feature set in the past operating environment sensor monitoring information. This matrix can be a two-dimensional matrix, with rows representing different fault event trigger features and columns representing different traceability viewpoints. For example, the fault event trigger features may include a sharp rise in temperature, a sudden change in humidity, a sudden increase in vibration amplitude, etc.

[0024] Taking the fault event trigger feature of a sharp rise in temperature as an example, in the first traceability discrimination thermal matrix, the element value corresponding to the traceability viewpoint of "heat sink blockage" may be 0.3 (indicating the possibility that the temperature rises sharply due to heat sink blockage), and the element value corresponding to "fan failure" may be 0.2, etc. This matrix comprehensively reflects the relationship between the fault event trigger feature and the traceability viewpoint, providing an important basis for subsequent model optimization.

[0025] Step 180: The environmental sensor analysis system optimizes the initial operation and maintenance fault tracing model based on the first tracing discrimination thermal matrix and the past operating environment sensor monitoring information to obtain a target operation and maintenance fault tracing model, wherein the target operation and maintenance fault tracing model is used to generate fault tracing viewpoints for servo controller fault events for the operating environment sensor monitoring information to be analyzed.

[0026] Finally, in step 180, the relevant steps of optimizing the target operation and maintenance fault tracing model are involved. The initial operation and maintenance fault tracing model can be a model built based on some basic algorithms, such as a rule-based fault tracing model, which may judge the cause of the fault according to some preset rules, such as "if the temperature exceeds 35°C and the humidity is less than 40%, it may be a heat dissipation problem". However, this model may have problems such as low accuracy and poor adaptability to complex situations.

[0027] When it comes to the optimization process, the environmental sensor analysis system optimizes the initial operation and maintenance fault tracing model based on the first tracing discrimination thermal matrix and the past operating environment sensor monitoring information. For example, a large amount of actual data in the past operating environment sensor monitoring information and the relationship information in the first tracing discrimination thermal matrix are used to optimize using a machine learning algorithm such as a neural network algorithm.

[0028] The environmental parameter values ​​and fault event triggering features in the past operating environment sensor monitoring information are used as the input of the neural network, and the corresponding fault tracing results (accurate results obtained based on the first tracing discrimination thermal matrix) are used as output for training. During the training process, the weights and other parameters of the neural network are adjusted so that the model can more accurately generate fault tracing opinions for servo controller fault events based on the input operating environment sensor monitoring information to be analyzed.

[0029] The optimized target operation and maintenance fault tracing model has higher accuracy and adaptability to complex environments. For example, when faced with new operating environment sensor monitoring information to be analyzed, it can more accurately determine the cause of the fault. For example, for a new combination of temperature, humidity and vibration data, it can accurately point out whether it is a specific component failure of the cooling system or a servo controller failure caused by environmental factors, thereby providing more effective support for the operation and maintenance of the servo controller.

[0030] It is worth mentioning that the present invention focuses on the sensor monitoring and fault tracing of the closed microclimate operating environment of the servo controller. Based on this perspective, the above steps 120 to 180 will be introduced and explained below.

[0031] (i) Step 120 involves obtaining prior environmental monitoring logs and past environmental monitoring logs - basic data acquisition for sensor monitoring and fault tracing.

[0032] Prior environment monitoring log: In the fault tracing system of the servo controller, the prior environment monitoring log is a critical part. It contains at least one prior operating environment sensor monitoring information and the corresponding servo controller fault event category characteristics.

[0033] The content of the prior operating environment sensor monitoring information is closely related to the closed microclimate operating environment. For example, for the environmental parameter of temperature, in a closed microclimate environment, the sensor may monitor with a certain accuracy (such as ±0.5°C). For example, a high-precision temperature sensor collects data every 5 minutes to form the time series data of the prior operating environment temperature sensor monitoring information.

[0034] For humidity, the sensor may measure with an accuracy of ±3%, and also record humidity values ​​at regular intervals (e.g., every 10 minutes). A vibration sensor may monitor vibration amplitude with a resolution of 0.01 mm / s and continuously collect data under different operating conditions.

[0035] The servo controller fault event category characteristics are clear identifications of different fault types. For example, in a closed microclimate environment, if poor heat dissipation causes the temperature to be too high, it may cause an "overheating fault". Its fault event category characteristics will record temperature-related abnormalities in detail, such as the length of time the temperature continues to be higher than the normal operating temperature upper limit (for example, the normal upper limit is 40°C, and the temperature reaches 50°C during the fault), the rate of temperature rise (such as an increase of 2°C per minute), etc. Similarly, humidity abnormalities may cause a "moisture short circuit fault", whose characteristics include the time and range of humidity exceeding the safety threshold (for example, the safety threshold is 70%, and it actually reaches 80%).

[0036] Past environment monitoring log: Past environment monitoring log contains multiple past operating environment sensor monitoring information. This information reflects the status of the closed microclimate operating environment over a period of time in the past.

[0037] Take temperature as an example. Past temperature sensing monitoring information may contain temperature data in different seasons and under different workloads. For example, during the hot summer period, the temperature in a closed microclimate environment may be generally high, and the data may fluctuate upward, with the highest temperature reaching 45°C. In winter, the temperature is relatively low, and may fluctuate between 10-20°C.

[0038] Humidity data also varies with the season and equipment operating conditions. In humid seasons, humidity may often approach or exceed 60%, while in dry seasons it remains at around 30-40%. Vibration data may be affected by equipment startup, shutdown, and load changes during operation. For example, at the moment of equipment startup, the vibration amplitude may suddenly increase to 0.1mm / s, and then decrease to below 0.05mm / s during stable operation. These past environmental monitoring logs provide a rich data foundation for subsequent fault tracing analysis.

[0039] (ii) Step 140 involves correlation analysis to obtain environmental factor impact assessment of past operating environment sensor monitoring information - closed microclimate environmental factors and fault correlation analysis.

[0040] The relationship between the fault performance variation involvement vector and the description semantics of the closed climate operating environment: The fault performance variation involvement vector is extracted from the prior operating environment sensor monitoring information, and is used to represent the quantitative characteristics of the deviation of environmental parameters from the normal state when a specific fault occurs. In a closed microclimate operating environment, for example, for an "overheating fault", the temperature component in the fault performance variation involvement vector may be expressed as the difference from the normal temperature range. For example, if the normal temperature range is 30-35°C, and the temperature reaches 45°C when the fault occurs, then the fault performance variation involvement vector value in the temperature dimension is +10°C (indicating a deviation of 10°C in the high temperature direction). For humidity, if the normal humidity range is 40-60%, and in a fault caused by abnormal humidity, the humidity reaches 80%, then the fault performance involvement vector value in the humidity dimension is +20%.

[0041] The closed climate operating environment description semantics is an intuitive description of the environmental state reflected by the past operating environment sensor monitoring information. For example, when the temperature is 40°C, the closed climate operating environment description semantics may be "high temperature state"; when the humidity is 70%, the semantics is "high humidity state"; when the vibration amplitude is 0.08mm / s, the semantics is "medium vibration state".

[0042] Association analysis algorithm example - Association analysis based on decision tree: When performing association analysis, a decision tree algorithm can be used. The root node of the decision tree can be the fault event category feature of the servo controller (such as "overheating fault", "short circuit fault due to moisture", "loose connection fault caused by vibration", etc.).

[0043] For the root node "overheating fault", the branch conditions of the decision tree can be based on the fault performance variation involvement vector of the prior operating environment sensor monitoring information and the closed climate operating environment description semantics of the past operating environment sensor monitoring information. For example, the branch condition of the temperature dimension can be: if the temperature value in the fault performance variation involvement vector is greater than +5°C (indicating a deviation of more than 5°C from the normal temperature range) and the closed climate operating environment description semantics in the past operating environment sensor monitoring information is "high temperature state" (temperature greater than 35°C), then continue to analyze the influence of other environmental factors such as humidity and vibration along this branch.

[0044] Through such decision tree analysis, the impact of various environmental factors is comprehensively considered to obtain the environmental factor impact evaluation of the past operating environment sensor monitoring information. For example, for a certain past operating environment sensor monitoring information, the environmental factor impact evaluation obtained after decision tree analysis can be: the impact degree of "overheating fault" is 0.6 (indicating that there is a 60% probability that it is related to overheating fault), the impact degree of "moisture short circuit fault" is 0.2, and the impact degree of "loose connection fault caused by vibration" is 0.2.

[0045] (iii) Step 160 involves fault tracing to obtain a fault tracing point of view thermal network and a first tracing discrimination thermal matrix - fault tracing based on closed microclimate environment analysis.

[0046] Construction of thermal network from the perspective of fault tracing - Taking temperature-related faults as an example: In a closed microclimate operating environment, a thermal network from the perspective of fault tracing is constructed by taking "overheating fault" as an example. Fault tracing perspectives may include failure of the internal cooling system of the equipment, poor ventilation in the closed microclimate environment, interference from external heat sources, etc.

[0047] The failure of the internal cooling system of the equipment may be related to factors such as the cooling efficiency of the heat sink and the circulation of the coolant. For example, if the cooling efficiency of the heat sink is reduced by 50% (the correlation strength with the fault can be obtained through experiments or theoretical calculations), then in the thermal network of the fault tracing viewpoint, the correlation strength between the reduced cooling efficiency of the heat sink and the "overheating fault" may be set to 0.4. For the poor ventilation of the closed microclimate environment, if the ventilation volume is less than 30% of the normal level (obtained through actual monitoring and analysis), its correlation strength with the "overheating fault" may be 0.3. In terms of external heat source interference, if the external heat source causes the temperature of the closed microclimate environment to rise by 10°C (based on actual measurement and analysis), its correlation strength with the "overheating fault" may be 0.2. These correlation strengths constitute the edge weights between nodes in the thermal network of the fault tracing viewpoint, reflecting the closeness between different fault tracing viewpoints and faults.

[0048] The construction and meaning of the first source tracing discriminant thermal matrix: The first source tracing discriminant thermal matrix is ​​constructed based on the fault tracing viewpoint thermal network. The rows of the matrix represent different fault event triggering features, and the columns represent different tracing viewpoints.

[0049] For example, the temperature in the fault event trigger feature rises sharply (for example, the temperature rises by more than 10°C within 10 minutes). In the first traceability discrimination thermal matrix, for the traceability point of view of internal cooling system failure of the equipment, the corresponding element value may be 0.3 (indicating that the possibility of the temperature rising sharply being caused by the internal cooling system failure of the equipment is 30%); for the closed microclimate environment with poor ventilation, the corresponding element value may be 0.25; for external heat source interference, the corresponding element value may be 0.15. This matrix comprehensively reflects the relationship between different fault event trigger features and fault tracing points of view in a closed microclimate operating environment, providing a quantitative basis for accurate fault tracing.

[0050] (iv) Step 180 involves optimizing the target operation and maintenance fault tracing model to improve the accuracy of fault tracing in a closed microclimate environment.

[0051] Limitations and improvement needs of the initial operation and maintenance fault tracing model: The initial operation and maintenance fault tracing model may have some limitations in a closed microclimate operating environment. For example, the initial model based on simple rules may only consider the impact of a single environmental factor on the servo controller failure, such as judging whether it is an overheating fault only based on whether the temperature exceeds the threshold, while ignoring humidity, vibration, and their interactions.

[0052] In the actual closed microclimate operating environment, multiple environmental factors are often interrelated. For example, high humidity may affect the heat dissipation efficiency, thereby indirectly causing overheating failures; vibration may loosen the heat dissipation components, affecting the heat dissipation effect and causing temperature anomalies. Therefore, it is necessary to optimize the initial model to improve its ability to handle complex fault conditions in a closed microclimate environment.

[0053] Optimization process - Genetic algorithm-based model optimization example: Genetic algorithm can be used to optimize the initial operation and maintenance fault tracing model. The element values ​​in the first tracing discriminant thermal matrix and the past operating environment sensor monitoring information are used as the input of the genetic algorithm.

[0054] In the genetic algorithm, the parameters of the model (such as the weights of different fault tracing viewpoints, the interaction coefficients between different environmental factors, etc.) are defined as chromosomes. Initially, a set of chromosomes is randomly generated to represent different combinations of model parameters. Then, these chromosomes are evaluated based on the fitness function (such as the degree of match between the fault tracing results predicted by the calculation model based on the past operating environment sensor monitoring information and the actual results).

[0055] Through genetic operations such as selection, crossover, and mutation, the chromosome population is continuously evolved, so that the value of the fitness function is continuously improved. For example, after multiple generations of evolution, a set of optimized model parameters is obtained, thereby obtaining a target operation and maintenance fault tracing model. This target model can more accurately consider the interaction between multiple environmental factors in a closed microclimate operating environment, and generate more accurate fault tracing viewpoints for servo controller fault events.

[0056] Through the above introduction to the key points of sensor monitoring and fault tracing of the servo controller monitoring the closed microclimate operating environment, we can have a deeper understanding of the meaning, operation method and logical relationship of each step in this specific environment, thereby providing more effective technical support for the reliable operation and fault maintenance of the servo controller in the closed microclimate operating environment.

[0057] First, the present invention can comprehensively integrate environmental sensor monitoring information and historical data related to fault category characteristics by acquiring prior environmental monitoring logs and past environmental monitoring logs, laying a solid foundation for subsequent analysis. The environmental factor impact evaluation obtained by correlating and analyzing the fault performance variation involvement vector of the prior operating environment sensor monitoring information with the closed climate operating environment description semantics of the past operating environment sensor monitoring information can accurately quantify the degree of influence of environmental factors on servo controller failures. Tracing the fault based on the environmental factor impact evaluation and constructing a fault tracing viewpoint thermal network can help to intuitively present various possible viewpoints of fault tracing and their correlation strengths, providing a comprehensive and intuitive perspective for fault tracing. The first tracing discrimination thermal matrix is ​​obtained from the fault tracing viewpoint thermal network. This matrix accurately reflects the relationship between the fault event triggering characteristics and the tracing viewpoint, providing a quantitative basis for fault tracing. Finally, the initial operation and maintenance fault tracing model is optimized by using the first traceability discrimination thermal matrix and past operating environment sensor monitoring information. The obtained target operation and maintenance fault tracing model greatly improves the accuracy of servo controller fault tracing, and can more effectively generate fault tracing opinions for the analyzed operating environment sensor monitoring information, thereby improving the reliability and maintenance efficiency of the servo control system, and providing strong technical support for intelligent operation and maintenance in the field of industrial automation.

[0058] In some optional embodiments, the fault tracing is performed based on the environmental factor impact evaluation to obtain the fault tracing viewpoint thermal network of the servo controller fault event in the past operating environment sensor monitoring information, including: capturing and identifying servo controller fault events based on the environmental factor impact evaluation to obtain the first abnormal performance feedback event thermal feature set of the past operating environment sensor monitoring information; performing a convolution operation on the first abnormal performance feedback event thermal feature set to obtain the dynamic performance mutation confidence weight of each first convolution kernel of the first abnormal performance feedback event thermal feature set; if the dynamic performance mutation confidence weight of the first convolution kernel is greater than a first set value, combining the first convolution feature corresponding to the dynamic performance mutation confidence weight of the first convolution kernel to determine the fault tracing viewpoint thermal network.

[0059] Furthermore, the servo controller fault data analysis method also includes: if the dynamic performance mutation confidence weights of the first convolution kernel are not greater than the first set value, then the past operating environment sensor monitoring information is used as the operating environment sensor monitoring information to be analyzed; the operating environment sensor monitoring information to be analyzed is fault traced through the target operation and maintenance fault tracing model to obtain the mask features of the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed.

[0060] In detail, the target operation and maintenance fault tracing model is used to perform fault tracing on the operating environment sensor monitoring information to be analyzed, and the mask features of the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed are obtained, which can be achieved through Example A and Example B.

[0061] Embodiment A: Through the target operation and maintenance fault tracing model, the operating environment sensor monitoring information to be analyzed is traced to obtain a second abnormal performance feedback event thermal feature set of the operating environment sensor monitoring information to be analyzed; a convolution operation is performed on the second abnormal performance feedback event thermal feature set to determine the dynamic performance mutation confidence weight of each second convolution kernel of the second abnormal performance feedback event thermal feature set; if the dynamic performance mutation confidence weight of the second convolution kernel is greater than a second set value, the second convolution feature corresponding to the dynamic performance mutation confidence weight of the second convolution kernel is used as a mask feature from the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed.

[0062] Embodiment B: Through the target operation and maintenance fault tracing model, the operating environment sensor monitoring information to be analyzed is traced to obtain a second abnormal performance feedback event thermal feature set of the operating environment sensor monitoring information to be analyzed; based on the second abnormal performance feedback event thermal feature set, the convolution feature with the largest abnormal performance feedback thermal coefficient is obtained as a mask feature from the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed.

[0063] In the present invention, the process of tracing the fault source based on the environmental factor impact assessment to obtain the thermal network fault tracing point of view is as follows.

[0064] (1) Capture and identification of servo controller fault events and acquisition of the thermal feature set of the first abnormal performance feedback event: In the process of tracing the fault source based on the environmental factor impact evaluation to obtain the fault tracing point of view thermal network, the servo controller fault event capture and identification is first performed. This process is based on the environmental factor impact evaluation obtained in the previous steps. For example, in the environmental factor impact evaluation, the impact weight of the temperature factor on the fault is 0.6 (indicating that the temperature factor accounts for a large proportion of the fault), the humidity factor has a weight of 0.3, and the vibration factor has a weight of 0.1. Based on these values ​​and the past operating environment sensor monitoring information, possible fault events are captured and identified.

[0065] In the recognition process, a pattern recognition algorithm, such as a classification algorithm based on a neural network, can be used. The input layer of the neural network receives parameters such as temperature, humidity, and vibration from past operating environment sensor monitoring information. The intermediate hidden layer processes this information through the connection weights between neurons, and the output layer outputs whether there is a fault event and the approximate type of the fault event. After this process, the thermal feature set of the first abnormal performance feedback event of the past operating environment sensor monitoring information is obtained. This feature set is a quantitative representation of performance feedback related to possible fault events. For example, for a fault that may be caused by excessive temperature, the temperature-related features in the thermal feature set of the first abnormal performance feedback event may be expressed as a value of the temperature deviating from the normal range (for example, the normal temperature range is 25-30°C, the actual temperature is 40°C, then the deviation value is 10°C) and the rate of temperature increase within a certain period of time (for example, an increase of 2°C per minute).

[0066] (2) Convolution operation and determination of confidence weight of dynamic performance mutation: After obtaining the thermal feature set of the first abnormal performance feedback event, a convolution operation is performed on it. The convolution operation is of great significance in this process, as it can extract local features of different parts of the feature set. For example, the thermal feature set of the first abnormal performance feedback event is a two-dimensional matrix (for example, rows represent different environmental parameter categories, and columns represent different time points or monitoring locations), and the convolution kernel can be a smaller matrix (such as a 3*3 matrix).

[0067] During the convolution process, the convolution kernel slides on the thermal feature set matrix of the first abnormal performance feedback event, and obtains a new value through the corresponding calculation rules (for example, multiplying and summing the corresponding elements), which is the convolution result. For each convolution kernel, such a convolution result can be obtained, which reflects the situation of different local characteristics. Then, the dynamic performance mutation confidence weight of each first convolution kernel is determined according to a certain algorithm. This algorithm can be based on statistical analysis methods, for example, calculating the difference between the convolution result and the convolution result under normal circumstances (no fault), and then determining the confidence weight according to the size and distribution of the difference.

[0068] For example, under normal circumstances, the mean of the convolution result is 10 and the standard deviation is 2. If the convolution result of a convolution kernel is 15, a relative deviation value is obtained through the corresponding calculation formula (such as (15-10) / 2=2.5), and then the dynamic performance mutation confidence weight is determined in the preset mapping relationship according to this deviation value (for example, when the deviation value is 2.5, the corresponding dynamic performance mutation confidence weight is 0.8).

[0069] (3) Determination of the fault tracing viewpoint thermal network: If the dynamic performance mutation confidence weight of the first convolution kernel is greater than the first set value (for example, the first set value is 0.6), the first convolution feature corresponding to the dynamic performance mutation confidence weight of the first convolution kernel is combined to determine the fault tracing viewpoint thermal network. For example, when the dynamic performance mutation confidence weight of a convolution kernel is 0.8 or greater than 0.6, the first convolution feature corresponding to this convolution kernel may be a feature related to poor heat dissipation caused by excessive temperature (such as excessive heat sink temperature, abnormal cooling fan speed, etc.). These features will be integrated into the fault tracing viewpoint thermal network. The fault tracing viewpoint thermal network is a network structure with fault tracing viewpoints as nodes and the strength of association between viewpoints as edge weights. In this example, there may be a certain strength of association between excessive heat sink temperature and abnormal cooling fan speed (for example, 0.4, indicating the possibility that both occur at the same time and cause a fault), and these associations will constitute the edge weights of the fault tracing viewpoint thermal network.

[0070] In addition, in the present invention, the processing when the confidence weight of the dynamic performance mutation of the first convolution kernel is not greater than the first set value is as follows: If the confidence weight of the dynamic performance mutation of the first convolution kernel is not greater than the first set value, the past operating environment sensor monitoring information is used as the operating environment sensor monitoring information to be analyzed. This means that in the previous convolution operation and confidence weight judgment, no sufficiently obvious fault features were found. At this time, the target operation and maintenance fault tracing model is used to trace the fault of the operating environment sensor monitoring information to be analyzed.

[0071] Fault tracing process of embodiment A

[0072] ① Obtain the thermal feature set of the second abnormal performance feedback event

[0073] Through the target operation and maintenance fault tracing model, the fault tracing of the operating environment sensor monitoring information to be analyzed is performed, and the second abnormal performance feedback event thermal feature set of the operating environment sensor monitoring information to be analyzed is first obtained. This process is similar to the previous process of obtaining the thermal feature set of the first abnormal performance feedback event, but is based on the target operation and maintenance fault tracing model. The target operation and maintenance fault tracing model may be a trained neural network model, whose input is the operating environment sensor monitoring information to be analyzed (including parameters such as temperature, humidity, vibration, etc.), and the output is a feature set related to the fault.

[0074] For example, the neural network structure of the target operation and maintenance fault tracing model is a multi-layer perceptron (MLP), with 3 nodes in the input layer corresponding to temperature, humidity, and vibration parameters, 5 nodes in the hidden layer, and 10 nodes in the output layer representing different fault-related features. Through the forward propagation algorithm, the input operating environment sensor monitoring information to be analyzed is propagated in the network to obtain the thermal feature set of the second abnormal performance feedback event.

[0075] ② Convolution operation and determination of confidence weights of dynamic performance mutation

[0076] A convolution operation is performed on the thermal feature set of the second abnormal performance feedback event. The convolution calculation is also performed using a convolution kernel (e.g., a 3*3 convolution kernel). For example, the thermal feature set of the second abnormal performance feedback event is a 10*10 matrix, and the convolution kernel slides on the matrix to calculate the convolution result.

[0077] Then determine the dynamic performance mutation confidence weight of each second convolution kernel of the thermal feature set of the second abnormal performance feedback event. This process is similar to the previous determination of the dynamic performance mutation confidence weight of the first convolution kernel, and the weight is also determined by comparing the difference between the convolution result and the normal situation. For example, using a method based on probability distribution, if the convolution result conforms to the probability distribution under a certain fault feature (such as normal distribution), the dynamic performance mutation confidence weight is determined according to the probability value.

[0078] ③ Determination of mask features from the perspective of fault tracing

[0079] If the dynamic performance mutation confidence weight of the second convolution kernel is greater than the second set value (for example, the second set value is 0.5), the second convolution feature corresponding to the dynamic performance mutation confidence weight of the second convolution kernel is used as the mask feature from the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed. For example, if the dynamic performance mutation confidence weight of a second convolution kernel is 0.7 and greater than 0.5, and its corresponding second convolution feature is a feature related to humidity sensor failure (such as abnormal fluctuations in humidity readings, etc.), then this feature is determined to be a mask feature from the fault tracing perspective. The mask feature can be used to highlight key features related to the fault in subsequent fault analysis and ignore other minor features.

[0080] Fault tracing process of embodiment B

[0081] ① Obtain the thermal feature set of the second abnormal performance feedback event

[0082] Similarly, the target operation and maintenance fault tracing model is used to trace the fault of the operating environment sensor monitoring information to be analyzed, and the thermal feature set of the second abnormal performance feedback event of the operating environment sensor monitoring information to be analyzed is obtained.

[0083] ②Determine the mask characteristics based on the thermal coefficient

[0084] According to the second abnormal performance feedback event thermal feature set, the convolution feature with the largest abnormal performance feedback thermal coefficient is obtained as the mask feature of the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed. For example, in the second abnormal performance feedback event thermal feature set, each convolution feature has a corresponding abnormal performance feedback thermal coefficient, which can reflect the degree of correlation between the feature and the fault. For example, through calculation, it is found that the abnormal performance feedback thermal coefficient of a certain convolution feature is 0.8, and the coefficients of other convolution features are all less than this value, then this convolution feature (which may be a feature related to loosening of components caused by vibration, such as abnormalities within a specific range of vibration frequency, etc.) is determined as a mask feature from the fault tracing perspective.

[0085] The technical solution of this embodiment firstly, through a series of operations of fault tracing based on the impact evaluation of environmental factors, including convolution operation and confidence weight judgment, it is possible to more accurately capture fault characteristics and construct a thermal network of fault tracing viewpoints, thereby improving the accuracy of fault tracing. In the convolution operation, through specific numerical calculations and weight determination, the fault-related features hidden in the data can be mined. Secondly, when the confidence weight of the dynamic performance mutation of the first convolution kernel does not meet the conditions, the target operation and maintenance fault tracing model is used for further analysis. Whether it is embodiment A or embodiment B, the mask features of the fault tracing viewpoint can be determined from different angles, which helps to further focus on the key factors of the fault and eliminate interference factors. This multi-level and multi-method fault tracing mechanism improves the reliability and flexibility of the entire servo controller fault analysis system, can better adapt to different fault conditions, reduce the time and cost of troubleshooting, and provide effective technical support for the operation and maintenance of servo controllers in the field of industrial automation.

[0086] In some preferred examples, the prior environment monitoring log includes first prior operating environment sensor monitoring information and second prior operating environment sensor monitoring information, and the fault performance variation involvement vector of the prior operating environment sensor monitoring information is correlated and analyzed with the closed climate operating environment description semantics of the past operating environment sensor monitoring information based on the servo controller fault event category characteristics to obtain the environmental factor impact evaluation of the past operating environment sensor monitoring information, including: obtaining the fault performance variation involvement vector of the first prior operating environment sensor monitoring information based on the servo controller fault event category characteristics of the first prior operating environment sensor monitoring information; obtaining the fault performance variation involvement vector of the second prior operating environment sensor monitoring information based on the servo controller fault event category characteristics of the second prior operating environment sensor monitoring information. quantity; semantic embedding is performed based on the past operating environment sensor monitoring information to obtain the closed climate operating environment description semantics of the past operating environment sensor monitoring information; correlation analysis is performed on the fault performance variation involvement vector of the first prior operating environment sensor monitoring information and the closed climate operating environment description semantics to obtain the first causal inference weight between the first prior operating environment sensor monitoring information and the past operating environment sensor monitoring information; correlation analysis is performed on the fault performance variation involvement vector of the second prior operating environment sensor monitoring information and the closed climate operating environment description semantics to obtain the second causal inference weight between the second prior operating environment sensor monitoring information and the past operating environment sensor monitoring information; the first causal inference weight and the second causal inference weight are weighted to obtain the environmental factor impact evaluation.

[0087] In detail, the servo controller fault event is a physical fault event, and the semantic embedding is performed based on the past operating environment sensor monitoring information to obtain the closed climate operating environment description semantics of the past operating environment sensor monitoring information, including: performing physical environment sensor feature mining on the past operating environment sensor monitoring information to obtain the physical environment sensor feature set of the past operating environment sensor monitoring information; performing semantic embedding on the physical environment sensor feature set to obtain the closed climate operating environment description semantics of the past operating environment sensor monitoring information.

[0088] It can be understood that the above embodiments focus on the process of obtaining the environmental factor impact assessment through association analysis.

[0089] (I) Obtaining the fault performance anomaly involvement vector of the first prior operating environment sensor monitoring information: In the present technical solution, the prior environment monitoring log includes the first prior operating environment sensor monitoring information and the second prior operating environment sensor monitoring information. For the first prior operating environment sensor monitoring information, the fault performance anomaly involvement vector is obtained based on its servo controller fault event category characteristics.

[0090] For example, if the servo controller fault event is an overheating fault (a physical fault event), in the first prior operating environment sensor monitoring information, if the temperature range is set to 20-30°C during normal operation, and the temperature reaches 40°C when the fault occurs, then in the temperature dimension, the fault performance variation involvement vector may be obtained by calculating the difference between the actual temperature and the upper limit of the normal temperature range, that is, 40-30=10°C, indicating a 10°C variation in the high temperature direction. Similar calculations can be performed for other relevant environmental parameters such as humidity and vibration. For example, the normal humidity range is 40%-60%, and the humidity is 70% during a fault. The variation involvement vector of the humidity dimension is 70-60=10%. The variation vectors of these different parameters are combined to form the fault performance variation involvement vector of the first prior operating environment sensor monitoring information.

[0091] (ii) Obtaining the fault performance variation involvement vector of the second prior operating environment sensor monitoring information: Similarly, obtaining the fault performance variation involvement vector of the servo controller based on the fault event category characteristics of the second prior operating environment sensor monitoring information.

[0092] For example, if the fault event is a short circuit fault due to moisture (also a physical fault event), in the second priori operating environment sensor monitoring information, for example, the normal humidity range is 40%-60%, and the humidity reaches 80% when the fault occurs. The fault performance variation involvement vector of the humidity dimension is 80-60=20%. For other factors that may affect the short circuit due to moisture, such as temperature (for example, the temperature is 25°C, the normal temperature range is 20-30°C, and the temperature does not exceed the normal range under this fault event, but it may have a certain impact), its component in the fault performance variation involvement vector can be determined according to a certain association algorithm (for example, through an empirical formula or a statistical model based on historical data, for example, it is concluded that the association weight of temperature to the short circuit fault due to moisture is 0.2, then the component of temperature in the fault performance variation involvement vector is (25-25)*0.2=0. The present invention calculates the difference based on the average temperature, and the algorithm can be adjusted according to the specific situation in practice). The vector components of each related parameter are combined to form the fault performance variation involvement vector of the second priori operating environment sensor monitoring information.

[0093] 3. Semantic description of closed climate operating environment to obtain past operating environment sensor monitoring information

[0094] ① Mining physical environment sensing feature sets

[0095] Since the servo controller failure event is a physical failure event, the physical environment sensing feature mining is performed on the past operating environment sensing monitoring information. For example, the past operating environment sensing monitoring information includes temperature, humidity, vibration and other data. For temperature data, its physical environment sensing features can be mined by analyzing the temperature fluctuation amplitude, change frequency and other features.

[0096] For example, if the temperature data sequence within a period of time is [22, 23, 25, 24, 26, 28] (unit: °C), the temperature fluctuation amplitude can be calculated, such as the sum of the absolute values ​​of the differences between adjacent data divided by the number of data, that is, ((|23-22|+|25-23|+|24-25|+|26-24|+|28-26|) / 6=1.33°C (the present invention is only an exemplary example of fluctuation amplitude calculation, and a more complex algorithm can be used in practice). This fluctuation amplitude is a physical environmental sensing feature of temperature. Similar feature mining is also performed on humidity and vibration data, such as the rate of change of humidity, the peak value of vibration, etc. These features are combined to obtain the physical environmental sensing feature set of past operating environment sensing monitoring information.

[0097] Semantic embedding obtains the semantics of closed climate operation environment description

[0098] The obtained physical environment sensing feature set is semantically embedded. The algorithm in the word vector model, such as the idea of ​​the Word2Vec algorithm, can be used for semantic embedding. Each feature in the physical environment sensing feature set is regarded as a "word" and a feature vector space is constructed.

[0099] For example, the vector corresponding to the temperature fluctuation of 1.33°C is [0.1, 0.2, -0.1] (this is the vector representation obtained through algorithm mapping, just an example), and the vector corresponding to the humidity change rate (for example, 0.5% / min) is [0.3, -0.1, 0.2], etc. Through the position relationship of these vectors in the feature vector space, the semantic representation of the past operating environment sensor monitoring information is obtained, that is, the closed climate operating environment description semantics. For example, when the temperature fluctuation is small, the humidity change rate is moderate, and the vibration peak is low, the closed climate operating environment description semantics may be "relatively stable operating environment".

[0100] (iv) Calculating the first causal inference weight: Performing correlation analysis on the fault performance anomaly involvement vector of the first a priori operating environment sensor monitoring information and the description semantics of the closed climate operating environment to obtain the first causal inference weight.

[0101] An algorithm based on logistic regression can be used for association analysis. Each component in the fault performance variation involvement vector of the first priori operating environment sensor monitoring information is used as the input feature of the logistic regression model, and the code corresponding to the description semantics of the closed climate operating environment (for example, "relatively stable operating environment" is encoded as 0, "unstable operating environment" is encoded as 1, etc.) is used as the target variable. For example, the fault performance variation involvement vector of the first priori operating environment sensor monitoring information is [10, 10%, 0.5] (temperature variation, humidity variation, vibration variation, respectively, and the vibration variation of 0.5 in the present invention is a relative value for example), and after calculation of the logistic regression model (the model parameters are obtained through historical data training, for example, the weight vector of the model is [0.2, 0.3, 0.1], and the intercept is -0.5), the logistic regression value is calculated to be 0.2*10+0.3*10%+0.1*0.5-0.5=1.85. This value is then mapped to between 0 and 1 through the sigmoid function to obtain the first causal inference weight. For example, the value calculated by the sigmoid function is 0.85.

[0102] (V) Calculating the second causal inference weight: Similar to the process of calculating the first causal inference weight, a correlation analysis is performed on the fault performance anomaly involvement vector of the second prior operating environment sensor monitoring information and the closed climate operating environment description semantics to obtain the second causal inference weight.

[0103] For example, the fault performance variation involvement vector of the second prior operating environment sensor monitoring information is [0, 20%, 0] (the present invention is the vector under the damp short circuit fault calculated previously), and the logistic regression value calculated by the same logistic regression model (the weight vector and intercept may be different, for example, the weight vector is [0.1, 0.4, 0.1], and the intercept is -0.3) is 0.1*0+0.4*20%+0.1*0-0.3=0.5. After the sigmoid function mapping, the second causal inference weight is obtained, for example, 0.62.

[0104] (VI) Obtaining the environmental factor impact evaluation: Weight the first causal deduction weight and the second causal deduction weight to obtain the environmental factor impact evaluation. For example, the weight of the first causal deduction weight is 0.6, and the weight of the second causal deduction weight is 0.4 (these weights can be obtained based on prior knowledge or historical data statistical analysis). Then the environmental factor impact evaluation is 0.6*0.85+0.4*0.62=0.758. This value can be used to represent the comprehensive impact of environmental factors in the past operating environment sensor monitoring information on servo controller failure events.

[0105] With this design, by respectively obtaining the fault performance variation involvement vectors of the first prior operating environment sensor monitoring information and the second prior operating environment sensor monitoring information, it is possible to comprehensively consider the variation of environmental factors related to different faults. When obtaining the closed climate operating environment description semantics of past operating environment sensor monitoring information, the operating environment is accurately described semantically from the process of mining physical environmental sensor features to semantic embedding. Algorithms such as logistic regression are used to calculate the causal deduction weights, and the environmental factor impact evaluation is obtained by weighting. This method effectively combines prior information with past information, and can accurately quantify the impact of environmental factors on servo controller fault events. This helps to improve the accuracy of fault tracing and provide a reliable basis for subsequent fault diagnosis and maintenance strategy formulation, thereby improving the reliability and maintenance efficiency of the servo control system and reducing production losses caused by faults.

[0106] In a replaceable embodiment, the initial operation and maintenance fault tracing model is optimized based on the first traceability discrimination thermal matrix and the past operating environment sensor monitoring information to obtain a target operation and maintenance fault tracing model, including: tracing the past operating environment sensor monitoring information through the initial operation and maintenance fault tracing model to obtain a first traceability prediction thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information; obtaining a first model training error of the initial operation and maintenance fault tracing model based on the first traceability prediction thermal matrix and the first traceability discrimination thermal matrix; optimizing the initial operation and maintenance fault tracing model based on the first model training error to obtain a target operation and maintenance fault tracing model.

[0107] On the one hand, the initial operation and maintenance fault tracing model is used to trace the faults of the past operating environment sensor monitoring information to obtain a first tracing prediction thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information, including: semantically embedding the past operating environment sensor monitoring information through the initial operation and maintenance fault tracing model to obtain the original environmental monitoring vector of the past operating environment sensor monitoring information; fully connecting the original environmental monitoring vector of the past operating environment sensor monitoring information through the initial operation and maintenance fault tracing model to obtain the environmental monitoring fully connected vector of the past operating environment sensor monitoring information; and performing feature upsampling based on the original environmental monitoring vector and the environmental monitoring fully connected vector to obtain the first tracing prediction thermal matrix of the past operating environment sensor monitoring information.

[0108] On the other hand, the initial operation and maintenance fault tracing model is optimized according to the first model training error to obtain a target operation and maintenance fault tracing model, including: obtaining a second tracing discrimination thermal matrix of the fault event triggering feature set in the prior operating environment sensor monitoring information according to the servo controller fault event category characteristics of the prior operating environment sensor monitoring information; tracing the fault of the prior operating environment sensor monitoring information through the initial operation and maintenance fault tracing model to obtain a second tracing prediction thermal matrix of the fault event triggering feature set in the prior operating environment sensor monitoring information; obtaining a second model training error of the initial operation and maintenance fault tracing model according to the second tracing prediction thermal matrix and the second tracing discrimination thermal matrix; optimizing the initial operation and maintenance fault tracing model according to the first model training error and the second model training error to obtain a target operation and maintenance fault tracing model.

[0109] The process of optimizing the initial operation and maintenance fault tracing model according to the first tracing determination thermal matrix and the past operating environment sensor monitoring information in the above embodiment is as follows:

[0110] 1. Obtaining the first source prediction thermal matrix

[0111] ① Semantic embedding to obtain the original environment monitoring vector

[0112] First, the past operating environment sensor monitoring information is semantically embedded through the initial operation and maintenance fault tracing model to obtain the original environmental monitoring vector of the past operating environment sensor monitoring information. Semantic embedding is a technology that maps raw data to a low-dimensional vector space. In this process, corresponding algorithms may be used, such as a word vector generation algorithm based on a neural network.

[0113] For example, the past operating environment sensor monitoring information includes data such as temperature, humidity and vibration. For temperature data, if the normal temperature range is 20-30°C, the actual temperature value is 25°C, the humidity is 40% (the normal range is 30-60%), and the vibration amplitude is 0.05mm / s (the normal range is 0-0.1mm / s). After the semantic embedding algorithm (for example, each parameter is mapped to a multidimensional vector space according to certain rules), for example, the temperature of 25°C is mapped to the vector [0.1, 0.2, -0.1] (the present invention is only an example vector, the actual vector dimension may be higher and the value is determined according to the specific algorithm), the humidity of 40% is mapped to the vector [-0.1, 0.3, 0.1], and the vibration amplitude of 0.05mm / s is mapped to the vector [0.2, -0.2, 0.3]. Combining these vectors gives us the original environmental monitoring vector of past operating environment sensor monitoring information, such as [0.1, 0.2, -0.1, -0.1, 0.3, 0.1, 0.2, -0.2, 0.3].

[0114] ② Fully connected operation to obtain the fully connected vector of environmental monitoring

[0115] Next, the original environmental monitoring vector of the past operating environment sensor monitoring information is fully connected through the initial operation and maintenance fault tracing model to obtain the environmental monitoring fully connected vector of the past operating environment sensor monitoring information. The full connection operation is a common operation in neural networks, and each neuron is connected to all neurons in the previous layer.

[0116] For example, this fully connected layer of the initial operation and maintenance fault tracing model has 10 neurons, and the original environmental monitoring vector is used as input. After the weight matrix of the fully connected layer (for example, the weight matrix is ​​a 9*10 matrix, and the values ​​are randomly initialized and adjusted through training), it is multiplied with the original environmental monitoring vector and then added with the bias vector (for example, the bias vector is a 10-dimensional [0.1, 0.2, ..., 1.0]), the environmental monitoring fully connected vector is obtained. For example, the original environmental monitoring vector is a 9-dimensional vector [0.1, 0.2, -0.1, -0.1, 0.3, 0.1, 0.2, -0.2, 0.3]. After the fully connected operation, a 10-dimensional environmental monitoring fully connected vector is obtained, such as [1.2, 0.8, -0.3, 0.5, 0.9, 1.1, 0.7, 0.4, 0.6, 1.0] (the numerical values ​​of the present invention are examples obtained based on calculations such as ).

[0117] ③ Feature upsampling to obtain the first source prediction thermal matrix

[0118] Then, feature upsampling is performed based on the original environment monitoring vector and the environment monitoring fully connected vector to obtain a first traceability prediction thermal matrix of past operating environment sensor monitoring information. Feature upsampling can use an interpolation algorithm, such as a bilinear interpolation algorithm.

[0119] For example, the features contained in the original environmental monitoring vector and the environmental monitoring fully connected vector are represented on a two-dimensional plane through a certain mapping relationship (the present invention is only an example representation for the convenience of understanding the upsampling process). The low-resolution features (the features represented by the original environmental monitoring vector and the environmental monitoring fully connected vector) are calculated in the horizontal and vertical directions according to the values ​​of the surrounding feature points through a bilinear interpolation algorithm, and the resolution of the features is gradually improved to finally obtain the first traceability prediction thermal matrix. For example, a 3*3 first traceability prediction thermal matrix is ​​obtained, and the element values ​​in the matrix represent the degree of correlation between the triggering features of different fault events, such as [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]] (the numerical values ​​of the present invention are examples).

[0120] (II) Obtaining the first model training error

[0121] ① Principle of calculating the first model training error

[0122] The first model training error of the initial operation and maintenance fault tracing model is obtained according to the first traceability prediction thermal matrix and the first traceability discrimination thermal matrix. The first traceability discrimination thermal matrix is ​​a known matrix that accurately reflects the relationship between the triggering characteristics of the fault event and the traceability viewpoint, and the first traceability prediction thermal matrix is ​​the prediction result obtained by the initial operation and maintenance fault tracing model.

[0123] The mean square error (MSE) algorithm can be used to calculate the first model training error. The calculation formula of the mean square error is:

[0124] MSE=frac{1}{n}sum_{i=1}^{n}(y_{i}-hat{y}_{i})^{2}.

[0125] Where y_{i} is the element value (true value) in the first-source discriminant thermal matrix, hat{y}_{i} is the corresponding element value (predicted value) in the first-source prediction thermal matrix, and n is the total number of matrix elements.

[0126] ② Specific calculation example

[0127] For example, the first tracing discriminant thermal matrix is ​​[[0.2, 0.3, 0.4], [0.5, 0.6, 0.7], [0.8, 0.9, 1.0]], and the first tracing prediction thermal matrix is ​​[[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]].

[0128] According to the mean square error formula, n=9, first calculate the square of the difference of each element:

[0129] (0.2-0.1)^{2}=0.01, (0.3-0.2)^{2}=0.01, (0.4-0.3)^{2}=0.01,

[0130] (0.5-0.4)^{2}=0.01, (0.6-0.5)^{2}=0.01, (0.7-0.6)^{2}=0.01,

[0131] (0.8-0.7)^{2}=0.01, (0.9-0.8)^{2}=0.01, (1.0-0.9)^{2}=0.01.

[0132] Then the sum of the squares of these differences is divided by n, that is, frac{0.01*9}{9}=0.01, so the first model training error is 0.01.

[0133] (III) Optimizing the initial operation and maintenance fault tracing model based on the first model training error

[0134] ① Obtain the second source tracing discriminant thermal matrix and the second source tracing prediction thermal matrix

[0135] According to the servo controller fault event category characteristics of the prior operating environment sensor monitoring information, the second traceability discrimination thermal matrix of the fault event trigger feature set in the prior operating environment sensor monitoring information is obtained. This process is similar to obtaining the first traceability discrimination thermal matrix, except that the data source is the prior operating environment sensor monitoring information.

[0136] Through the initial operation and maintenance fault tracing model, the prior operating environment sensor monitoring information is traced to the fault source, and the second tracing prediction thermal matrix of the fault event triggering feature set in the prior operating environment sensor monitoring information is obtained. This process is also similar to obtaining the first tracing prediction thermal matrix, including steps such as semantic embedding, full connection operation and feature upsampling.

[0137] ②Calculate the second model training error

[0138] -According to the second traceability prediction thermal matrix and the second traceability discrimination thermal matrix, obtain the second model training error of the initial operation and maintenance fault tracing model. The mean square error algorithm is also used for calculation. For example, the second traceability discrimination thermal matrix is ​​[[0.3, 0.4, 0.5], [0.6, 0.7, 0.8], [0.9, 1.0, 1.1]], and the second traceability prediction thermal matrix is ​​[[0.2, 0.3, 0.4], [0.5, 0.6, 0.7], [0.8, 0.9, 1.0]]. The second model training error is calculated according to the mean square error formula (the calculation process is similar to the first model training error). For example, the second model training error is 0.02.

[0139] ③Final model optimization

[0140] According to the first model training error and the second model training error, the initial operation and maintenance fault tracing model is optimized to obtain the target operation and maintenance fault tracing model. The gradient descent algorithm can be used for optimization.

[0141] For example, the parameter of the initial operation and maintenance fault tracing model is theta, and the loss function is L=alpha*first model training error+beta*second model training error (where alpha and beta are weights, for example, alpha=0.6, beta=0.4). According to the gradient descent formula theta=theta-eta*nablaL (where eta is the learning rate, for example, eta=0.01), the gradient nablaL of the loss function L with respect to the parameter theta is calculated, and then the parameter theta is updated according to the formula. After multiple iterations (for example, 100 iterations), until the model converges, the optimized target operation and maintenance fault tracing model is obtained.

[0142] It can be seen that the above embodiment uses the initial operation and maintenance fault tracing model to operate the past operating environment sensor monitoring information to obtain the first tracing prediction thermal matrix, and compares it with the first tracing discrimination thermal matrix to calculate the first model training error, and at the same time performs similar operations on the prior operating environment sensor monitoring information to obtain the second model training error. This multi-faceted error calculation comprehensively evaluates the accuracy of the model. Specific algorithms such as mean square error are used to accurately quantify the error, and then the initial model is optimized based on the two model training errors through optimization algorithms such as gradient descent to obtain the target operation and maintenance fault tracing model. This process improves the accuracy of the fault tracing model, enabling it to more accurately trace the triggering features of the fault event, thereby improving the reliability and efficiency of fault diagnosis of the servo control system and reducing the operation and maintenance costs and production losses caused by inaccurate fault tracing.

[0143] Under some technical ideas, the first tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information is obtained based on the fault tracing viewpoint thermal network, including: quantizing the past operating environment sensor monitoring information according to the fault tracing viewpoint thermal network to obtain the quantized monitoring vector set of the past operating environment sensor monitoring information; performing thermal space migration transformation on the quantized monitoring vector set to obtain the first tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information.

[0144] In this technical idea, the process of obtaining the first tracing discrimination thermal matrix from the thermal network based on the fault tracing point of view is as follows.

[0145] 1. Quantitative processing of past operating environment sensor monitoring information to obtain a quantitative monitoring vector set

[0146] ①Principles and basis of quantitative processing

[0147] In the process of quantizing the past operating environment sensor monitoring information to obtain a quantitative monitoring vector set based on the fault tracing point of view thermal network, it is necessary to set the information structure and related parameters based on the fault tracing point of view thermal network. The fault tracing point of view thermal network is a network structure with fault tracing points of view as nodes and the strength of association between points of view as edge weights.

[0148] The past operating environment sensor monitoring information contains multiple environmental parameters, such as temperature, humidity, vibration and other data. For temperature data, the quantification method may be determined based on the fault tracing point of view related to temperature in the fault tracing point of view thermal network and its correlation strength. For example, in the fault tracing point of view thermal network, too high temperature is strongly associated with the "overheating fault" point of view, with a correlation strength of 0.8 (between 0-1 represents the degree of correlation), and the normal temperature range is set to 20-30°C. If the actual temperature is 35°C, which is beyond the normal range, the quantification process can be calculated based on the degree of excess and the strength of association with the "overheating fault".

[0149] ②Specific quantitative calculation examples

[0150] For temperature, calculate the proportion exceeding the upper limit of the normal range, that is, (35-30) / (30-20)=0.5. Then, combined with the correlation strength of 0.8 with "overheating fault", the quantified value of temperature is 0.5*0.8=0.4. For humidity data, for example, the normal humidity range is 40%-60%, the actual humidity is 70%, and the proportion exceeding the upper limit of the normal range is (70-60) / (60-40)=0.5. If in the thermal network from the fault tracing point of view, the correlation strength between excessive humidity and "moisture short circuit fault" is 0.6, then the quantified value of humidity is 0.5*0.6=0.3. Similarly, for vibration data, for example, the normal vibration amplitude is 0-0.1mm / s, the actual vibration amplitude is 0.15mm / s, and the proportion exceeding the upper limit of the normal range is (0.15-0.1) / (0.1-0)=0.5. If the correlation strength between excessive vibration and "loose component fault" in the thermal network from the fault tracing point of view is 0.7, then the quantized value of vibration is 0.5*0.7=0.35. Combining these quantized values, we can get a quantized monitoring vector set of past operating environment sensor monitoring information, such as [0.4, 0.3, 0.35] (the present invention is only an example, and the actual vector dimension and value are determined according to more environmental parameters and fault tracing points of view).

[0151] (II) Perform thermal space migration transformation on the quantitative monitoring vector set to obtain the first source tracing discriminant thermal matrix

[0152] ① Concept and purpose of thermal space migration transformation

[0153] Thermal space migration transformation is a specific transformation method that converts the quantitative monitoring vector set into the first traceability discriminant thermal matrix. Its purpose is to convert the quantified environmental parameter information into a matrix form that can accurately reflect the relationship between the fault event triggering feature set and the traceability viewpoint. This matrix form can be better used for subsequent fault tracing analysis.

[0154] In the thermal space, each dimension may represent a different fault event trigger feature, and the element values ​​in the matrix represent the degree of association or probability relationship between these trigger features and the traceability viewpoint.

[0155] ②Example of thermal space migration transformation based on mapping function

[0156] For example, there is a mapping function f(x) used for thermal space migration transformation, where x is an element in the quantitative monitoring vector set. For the quantitative monitoring vector set [0.4, 0.3, 0.35], the first element 0.4 is transformed. For example, the mapping function f(x) = frac{1}{1+e^{-x}} (this is a mapping function similar to the sigmoid function, which is used to map values ​​between 0-1 and has certain nonlinear characteristics), then f(0.4) = frac{1}{1+e^{-0.4}}approx0.6.

[0157] Similarly, transform the second element 0.3:

[0158] f(0.3)=frac{1}{1+e^{-0.3}}approx0.57, transform the third element 0.35, f(0.35)=frac{1}{1+e^{-0.35}}approx0.59. These transformed values ​​are constructed into a first traceability discrimination thermal matrix according to certain rules. For example, if the temperature-related fault event trigger features are placed in the first row, the humidity-related ones are placed in the second row, and the vibration-related ones are placed in the third row, a 3*3 first traceability discrimination thermal matrix is ​​constructed (in order to simplify the examples of the present invention, the actual matrix size is determined according to the number of fault event trigger features), and the following is obtained:

[0159] begin{bmatrix}0.6&0.1&0.10.1&0.57&0.10.1&0.1&0.59end{bmatrix}.

[0160] The non-diagonal elements in the matrix of the present invention are initially set to 0.1 (representing the initial correlation degree between the triggering features of different types of fault events, for example, values, which may be adjusted according to more prior knowledge in practice), and the diagonal elements are numerical values ​​transformed by the mapping function, indicating the correlation degree between the triggering features of each fault event and their own related traceability viewpoints.

[0161] In this way, by quantizing the past operating environment sensor monitoring information based on the thermal network of the fault tracing point of view, a quantitative monitoring vector set is obtained. This quantization method combines the correlation information in the thermal network of the fault tracing point of view, and can accurately convert the environmental parameter information into a quantitative representation related to fault tracing. Then, the quantized monitoring vector set is transformed into a thermal space migration to obtain the first traceability discrimination thermal matrix, which can clearly reflect the relationship between the fault event triggering feature set and the traceability point of view. The use of specific calculation methods such as calculating the quantization value based on the correlation strength and performing the thermal space migration transformation of the corresponding mapping function improves the accuracy and effectiveness of the fault tracing analysis, provides a reliable basis for subsequent fault diagnosis and operation and maintenance, and helps to reduce the time and cost of troubleshooting and improve the reliability of the system.

[0162] Under some innovative derivative design ideas, after optimizing the initial operation and maintenance fault tracing model according to the first tracing discrimination thermal matrix and the past operating environment sensor monitoring information to obtain the target operation and maintenance fault tracing model, the method also includes: outputting the current fault tracing viewpoint of the servo controller fault event of the operating environment sensor monitoring information to be analyzed through the target operation and maintenance fault tracing model; and adjusting the climate parameters of the closed environment based on the current fault tracing viewpoint.

[0163] First, the process of fault tracing and environmental parameter adjustment based on the target operation and maintenance fault tracing model is as follows.

[0164] (I) Current fault tracing perspectives of servo controller fault events that output sensor monitoring information of the operating environment to be analyzed through the target operation and maintenance fault tracing model

[0165] ① Input and processing of target operation and maintenance fault tracing model

[0166] The target operation and maintenance fault tracing model is optimized to more accurately trace the fault events of the servo controller. When processing the sensor monitoring information of the operating environment to be analyzed, the sensor monitoring information of the operating environment to be analyzed is first input into the target operation and maintenance fault tracing model. The sensor monitoring information of the operating environment to be analyzed contains various parameters related to the operating environment of the servo controller, such as temperature, humidity, vibration and other information.

[0167] For example, the temperature value in the operating environment sensor monitoring information to be analyzed is 32°C, the humidity is 55%, and the vibration amplitude is 0.08mm / s. These parameters are input into the target operation and maintenance fault tracing model in the corresponding format, and the model may process these inputs based on the features and relationships learned in the previous training process.

[0168] ② Output the specific calculation and analysis of the current fault tracing point of view

[0169] In the target operation and maintenance fault tracing model, a neural network structure, such as a multi-layer perceptron (MLP), may be used. The input layer receives the various parameters of the operating environment sensor monitoring information to be analyzed, and performs complex nonlinear transformations through the neurons in the hidden layer. For example, the input layer has 3 nodes corresponding to temperature, humidity, and vibration, the hidden layer has 5 neurons, and the output layer has several nodes representing different fault tracing viewpoints.

[0170] The weights and biases of the model are determined during the optimization process. For the input temperature of 32°C, humidity of 55%, and vibration amplitude of 0.08mm / s, after calculation of the hidden layer (for example, the activation function of the hidden layer neuron is the ReLU function, and the calculation process involves matrix multiplication and activation function operations), the probability value or score of each fault tracing viewpoint is finally obtained in the output layer. For example, the node output value of the output layer representing "overheating fault" is 0.2, the node output value of "moisture short circuit fault" is 0.1, and the node output value of "vibration causes loose components fault" is 0.3 (the numerical values ​​of the present invention are only examples, and are actually calculated based on the model). Based on these output values, it can be determined that the current fault tracing viewpoint is "vibration causes loose components fault" because its output value is the highest.

[0171] 2. Adjustment of climate parameters in closed environment based on current fault tracing perspectives

[0172] ① Climate parameter adjustment strategies for different fault tracing viewpoints

[0173] When the current fault tracing point of view is determined to be "vibration causes loose components", the climate parameters of the closed environment need to be adjusted. For this fault tracing point of view, possible adjustment strategies are related to environmental factors related to vibration. For example, if it is found that the vibration is caused by thermal expansion and contraction caused by changes in ambient temperature, which leads to loose components, then the climate parameter of temperature needs to be adjusted.

[0174] For example, under normal circumstances, the temperature fluctuation range of a closed environment is allowed to be between 20-30°C, but since the current fault tracing view is "vibration causes loose components" and is related to temperature, it may be necessary to adjust the temperature fluctuation range to a narrower range, such as 22-28°C. This can be achieved by adjusting the temperature control system of the closed environment. For example, if the temperature is adjusted through the air conditioning system, the temperature control parameters of the air conditioning need to be reset to set the target temperature range to 22-28°C.

[0175] If the current fault tracing view is "short circuit fault due to moisture", the climate parameter of humidity needs to be adjusted. For example, the normal humidity range is 30%-60%. In order to avoid short circuit fault due to moisture, the humidity may need to be reduced to the range of 30%-40%. This can be achieved through dehumidification equipment, adjusting the operating parameters of the dehumidification equipment, such as increasing the dehumidification power or adjusting the dehumidification cycle.

[0176] ② Specific algorithms and operation examples for climate parameter adjustment

[0177] For temperature adjustment, if the PID (proportional-integral-differential) control algorithm is used to adjust the temperature control system. For example, if the current temperature is 32°C and the target temperature range is 22-28°C, the parameters of the PID algorithm are K_p=0.5, K_i=0.1, K_d=0.2 (these parameters are determined based on system characteristics and experience). First, calculate the error e=32-28=4 (take the difference from the upper limit of the target range as the error).

[0178] Proportional term P = K_p*e = 0.5*4 = 2, integral term I = K_i*int_{0}^{t}edt (for example, integral calculation over a period of time, the simplified example of the present invention does not calculate the integral term in detail), differential term D = K_d*frac{de}{dt} (for example, the temperature change rate is 1°C / minute, then D = 0.2*1 = 0.2). Then, the output power of the temperature control system is adjusted according to the output u = P + I + D of the PID algorithm (the simplified calculation result of the present invention is u = 2 + 0 + 0.2 = 2.2, and the exact value is determined according to the complete integral calculation and system characteristics in practice), thereby achieving temperature adjustment.

[0179] For humidity adjustment, if a humidity sensor is used to monitor humidity in real time, when the humidity is higher than the upper limit of the target range (such as 40%), the dehumidification equipment starts to work. For example, the dehumidification efficiency of the dehumidification equipment is to reduce the humidity by 10% per hour (this is determined according to the performance of the dehumidification equipment). If the current humidity is 50%, then after 1 hour of dehumidification operation, the humidity can theoretically be reduced to 40%, reaching the target range.

[0180] In this way, the current fault tracing viewpoint of the servo controller fault event of the operating environment sensor monitoring information to be analyzed can be accurately output through the target operation and maintenance fault tracing model, and the possible cause of the fault can be accurately located. Then, based on the current fault tracing viewpoint, the climate parameters of the closed environment are adjusted, and specific algorithms such as PID control algorithm are used to adjust the temperature and adjust the humidity according to the performance of the dehumidification equipment. This targeted adjustment strategy helps prevent and solve servo controller failures. It not only improves the reliability and stability of the servo controller and reduces the frequency of failures, but also extends the service life of the equipment, reduces operation and maintenance costs, and improves the operating efficiency of the entire system.

[0181] Under some innovative derivative design ideas, the method also includes: correlating the operating environment sensor monitoring information to be analyzed, the current fault tracing viewpoint and the climate parameter adjustment record for disaster recovery processing.

[0182] In detail, the step of associating the to-be-analyzed operating environment sensor monitoring information, the current fault tracing viewpoint, and the climate parameter adjustment record for disaster recovery may include the following contents:

[0183] Data preparation step: Obtain the sensor monitoring information of the operating environment to be analyzed, the current fault tracing point of view, and the climate parameter adjustment record, and use them as the data combination to be processed for associated disaster recovery;

[0184] Data preprocessing steps: Standardize the operating environment sensor monitoring information to be analyzed, and unify the environmental parameter data of different types, different precisions and formats into a standard format, where environmental parameters include but are not limited to temperature, humidity, vibration and other data; convert the current fault tracing point of view into a standard coding form, and encode it through a predefined fault tracing point of view coding dictionary; standardize the climate parameter adjustment records, and record the timestamp of each adjustment, the name of the adjusted parameter, the value before adjustment, the target value after adjustment, and the adjustment method in a fixed format;

[0185] The step of establishing the association relationship is to establish the association relationship between the above three by establishing an index table. The records in the index table correspond to the operating environment sensor monitoring information instance to be analyzed. The records contain the unique identifier of the instance, the corresponding current fault tracing point of view code, and the pointer to the climate parameter adjustment record related to the current fault tracing. Alternatively, the association rule algorithm in data mining is used, such as the Apriori algorithm, to mine the potential association relationship between the environmental parameters in the operating environment sensor monitoring information to be analyzed and the current fault tracing point of view, as well as the association relationship between the current fault tracing point of view and the climate parameter adjustment record.

[0186] Data storage and backup steps: adopt a hierarchical storage and backup strategy for the data after correlation processing, store recent data in high-speed, easily accessible storage media, store mid-term data in large-capacity disk arrays, and back up long-term data off-site; use encryption technology such as AES algorithm to encrypt and protect data during data storage;

[0187] Disaster recovery steps: When local data is damaged, use the mid-term data stored in the disk array for recovery, find the relevant complete information through the index table or association rules, and repair and rebuild it in a standardized format; when the entire local storage system fails, use the off-site backup data for recovery, first decrypt the off-site backup data, then restore the data to the new storage system according to the storage structure and association relationship, and use the association rules to verify and optimize the restored data;

[0188] Data utilization steps: Use historical data after associated disaster recovery processing to optimize the target operation and maintenance fault tracing model, discover potential defects in the model by analyzing the associated relationship and adjust the model parameters; use associated data for predictive maintenance, predict the type of faults that may occur under specific environmental parameter conditions based on historical associated relationships and take measures in advance.

[0189] Furthermore, the above-mentioned disaster recovery processing technical solution is described as follows.

[0190] 1. Basic concepts and purposes of disaster recovery

[0191] Correlated disaster recovery processing is another innovative design of the present invention based on fault tracing and environmental parameter adjustment. Its purpose is to organically combine and process the operating environment sensor monitoring information to be analyzed, the current fault tracing viewpoint, and the climate parameter adjustment record to cope with various possible disaster situations (such as data loss, equipment failure, etc.), while providing a more comprehensive and valuable data foundation for subsequent operation and maintenance, optimization, and fault prevention. This processing method is not a simple data storage or recording, but a deep correlation integration, which aims to improve the reliability, recoverability, and fault prevention capabilities of the system from multiple dimensions.

[0192] (II) Specific operation process of associated disaster recovery

[0193] ①Data standardization and formatting

[0194] First, the operating environment sensor monitoring information to be analyzed, the current fault tracing viewpoint, and the climate parameter adjustment records are standardized. For the operating environment sensor monitoring information to be analyzed, it contains various types of environmental parameter data, such as temperature, humidity, vibration, etc. These data may come from different sensors and have different precision and data formats. For example, temperature data may be in degrees Celsius, accurate to one decimal place; humidity data may be expressed as a percentage, accurate to an integer; vibration data may be expressed in corresponding physical quantities (such as mm / s), with different ranges and precisions. These data need to be unified into a standard format for subsequent association processing.

[0195] For the current fault tracing point of view, it is converted into a standard coding form. For example, there is a predefined fault tracing point of view coding dictionary, in which "overheating fault" is coded as 001, "short circuit fault due to moisture" is coded as 002, "loose components caused by vibration" is coded as 003, etc. In this way, the fault tracing point of view can be expressed in a unified form that is easy for computer processing.

[0196] Climate parameter adjustment records also need to be standardized. For example, record the timestamp of each adjustment, the name of the adjusted parameter (such as temperature, humidity, etc.), the value before adjustment, the target value after adjustment, and the adjustment method (such as manual adjustment, automatic adjustment equipment, etc.). This information is stored in a fixed format, such as XML or JSON format, to ensure data consistency and scalability.

[0197] ②Establishment of related relationships

[0198] After data standardization, the association between the three begins. One way to achieve association is by establishing an index table. Each record in the index table corresponds to a corresponding instance of operating environment sensor monitoring information to be analyzed. The index table contains the unique identifier of the instance (such as a timestamp or data acquisition sequence number), the current fault tracing point of view code corresponding to the instance, and the pointer to the climate parameter adjustment record related to the fault tracing (which can be a file storage location or an index of a database record).

[0199] Another way to establish association relationships is to use association rule algorithms in data mining. For example, the Apriori algorithm can be used to mine the potential association relationship between the environmental parameters in the operating environment sensor monitoring information to be analyzed and the current fault tracing point of view. For example, after analysis, it was found that when the temperature is higher than 35°C and the humidity is lower than 30%, the probability of "overheating fault" (coded as 001) as the current fault tracing point of view is higher. At the same time, it was found that when an "overheating fault" occurs, the climate parameter adjustment record often contains a temperature reduction operation. Through this kind of association rule mining, we can have a deeper understanding of the internal connection between the three, and provide richer knowledge for subsequent disaster recovery processing.

[0200] ③Data storage and backup strategy

[0201] For the data after correlation processing, a hierarchical storage and backup strategy is adopted. The recent (such as the last month) correlation data is stored in a high-speed, easily accessible storage medium, such as a solid-state drive (SSD). This part of the data is most likely to be frequently accessed and used for real-time analysis, such as for short-term fault review or performance evaluation.

[0202] Medium-term (e.g. 1-6 months) data is stored in large-capacity disk arrays. Although this data is accessed relatively infrequently, it still has important reference value, such as for seasonal failure analysis or long-term trend research.

[0203] For long-term data (such as more than 6 months), off-site backup is performed. You can choose to back up the data to a remote data center or cloud storage service. Off-site backup can prevent data loss in the event of a natural disaster, fire or other unforeseen disaster in the local area. During data storage, encryption technology is used to protect the data to ensure data security and integrity. For example, the AES (Advanced Encryption Standard) algorithm is used to encrypt the stored data, and only the authorized key can decrypt and access the data.

[0204] ④Disaster recovery mechanism

[0205] In the event of a disaster (such as data corruption, storage device failure, etc.), the disaster recovery mechanism is activated. If the data is partially damaged, such as partial damage to recent data in the SSD, it can be recovered using the mid-term data stored in the disk array. Through the index table or association rules, complete information related to the damaged data can be found, and then repaired and rebuilt according to the standardized format of the data.

[0206] If the entire local storage system fails (such as a disk array failure), the off-site backup data is used for recovery. During the recovery process, the off-site backup data must first be decrypted (if it is encrypted), and then the data is gradually restored to the new storage system according to the data storage structure and association relationship. At the same time, the association rules obtained by data mining are used to verify and optimize the recovered data to ensure the accuracy and integrity of the data. For example, if it is found that some associations in the recovered data do not conform to the rules obtained by previous mining (such as an abnormality in the association between temperature and fault tracing), it is necessary to further check the accuracy of the data or re-perform data mining analysis.

[0207] ⑤Data utilization based on associated disaster recovery processing

[0208] In the normal operation and maintenance process, the data after the associated disaster recovery processing also has important utilization value. For example, the historical associated data can be used to further optimize the target operation and maintenance fault tracing model. By analyzing the association between a large amount of operating environment sensor monitoring information to be analyzed, the current fault tracing viewpoints, and the climate parameter adjustment records, potential defects or inaccuracies in the model can be discovered, so as to adjust the parameters of the model.

[0209] At the same time, the correlation data can also be used for predictive maintenance. Based on the correlation relationship in the historical data, the types of faults that may occur under the corresponding environmental parameter conditions can be predicted, and corresponding climate parameter adjustment measures or preventive maintenance operations can be taken in advance. For example, if it is found based on historical correlation data that the trend of temperature and humidity changes in the upcoming season may lead to an increase in the probability of "short circuit faults caused by moisture", then the humidity can be controlled in advance or the equipment can be treated with moisture-proof treatment.

[0210] It should be noted that the innovation of the above-mentioned embodiments is reflected in the deep association of seemingly independent operating environment sensor monitoring information to be analyzed, current fault tracing viewpoints, and climate parameter adjustment records. This association method has not been fully valued in traditional servo controller fault handling. Through the associated disaster recovery processing, data standardization, mining of associated relationships, and hierarchical storage and backup are achieved, which improves the manageability and security of data. In terms of disaster recovery, whether it is local data damage or failure of the entire storage system, data recovery can be effectively performed to ensure the continuous operation of the system. In addition, the data after the associated disaster recovery processing can also be used for model optimization and predictive maintenance, further improving the reliability and fault prevention capabilities of the system. This is a non-obvious innovation that provides a more comprehensive and efficient solution for the operation and maintenance of servo controllers.

[0211] Further, Figure 2 FIG. 2 is a schematic diagram of the structure of an environmental sensing analysis system 200 provided by the present invention. Figure 2 The environment sensing and analysis system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method of the present invention. Figure 2 As shown, the environment sensing analysis system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method of the present invention. The memory 230 may be a separate device independent of the processor 210, or may be integrated in the processor 210. Optionally, as Figure 2As shown, the environmental sensing analysis system 200 may also include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. Optionally, the environmental sensing analysis system 200 may implement the corresponding processes corresponding to the storage engine or the components (such as processing modules) in the storage engine or the device deployed with the storage engine in each method of the present invention, and for the sake of brevity, it will not be repeated here. It should be understood that the processor of the present invention may be an integrated circuit chip with signal processing capabilities. It can be understood that the memory in the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. It should be noted that the memory of the system and method described herein is intended to include but is not limited to a suitable type of memory. On the basis of the above, a readable storage medium is provided, on which a program or instruction is stored, and the steps of the above method are implemented when the program or instruction is executed by the processor.

[0212] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in the field shall fall within the protection of the present invention without departing from the purpose of the present invention and the scope of protection of the present invention.

Claims

1. A servo controller fault data analysis method, characterized in that: The servo controller fault data analysis method is applied to an environmental sensing analysis system, and the servo controller fault data analysis method includes: Acquire a priori environment monitoring log and a past environment monitoring log, wherein the priori environment monitoring log includes at least one priori operating environment sensor monitoring information and a servo controller fault event category feature of the priori operating environment sensor monitoring information, and the past environment monitoring log includes a plurality of past operating environment sensor monitoring information; According to the servo controller fault event category characteristics, the fault performance variation involvement vector of the prior operating environment sensor monitoring information and the closed climate operating environment description semantics of the past operating environment sensor monitoring information are correlated and analyzed to obtain the environmental factor impact evaluation of the past operating environment sensor monitoring information; Perform fault tracing based on the environmental factor impact evaluation to obtain the fault tracing viewpoint of the servo controller fault event in the past operating environment sensor monitoring information; According to the fault tracing viewpoint thermal network, a first tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information is obtained; Based on the first traceability discrimination thermal matrix and the past operating environment sensor monitoring information, the initial operation and maintenance fault tracing model is optimized to obtain a target operation and maintenance fault tracing model, wherein the target operation and maintenance fault tracing model is used to generate fault tracing viewpoints for servo controller fault events for the operating environment sensor monitoring information to be analyzed.

2. The servo controller fault data analysis method according to claim 1, characterized in that: The fault tracing is performed based on the environmental factor impact evaluation to obtain the fault tracing point of view of the servo controller fault event in the past operating environment sensor monitoring information, including: Capturing and identifying servo controller fault events based on the environmental factor impact evaluation to obtain a thermal feature set of a first abnormal performance feedback event of the past operating environment sensor monitoring information; Performing a convolution operation on the first abnormal performance feedback event thermal feature set to obtain a dynamic performance mutation confidence weight of each first convolution kernel of the first abnormal performance feedback event thermal feature set; If the dynamic performance mutation confidence weight of the first convolution kernel is greater than a first set value, the fault tracing point of view thermal network is determined by combining the first convolution feature corresponding to the dynamic performance mutation confidence weight of the first convolution kernel.

3. The servo controller fault data analysis method according to claim 2, characterized in that: The servo controller fault data analysis method also includes: If the dynamic performance mutation confidence weights of the first convolution kernel are not greater than the first set value, the past operating environment sensor monitoring information is used as the operating environment sensor monitoring information to be analyzed; The target operation and maintenance fault tracing model is used to perform fault tracing on the operating environment sensor monitoring information to be analyzed, and a mask feature of the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed is obtained.

4. The servo controller fault data analysis method according to claim 3, characterized in that: The method of performing fault tracing on the operating environment sensor monitoring information to be analyzed by using the target operation and maintenance fault tracing model to obtain a mask feature of the fault tracing viewpoint of the servo controller fault event in the operating environment sensor monitoring information to be analyzed includes: Through the target operation and maintenance fault tracing model, the fault tracing is performed on the operating environment sensor monitoring information to be analyzed to obtain a second abnormal performance feedback event thermal feature set of the operating environment sensor monitoring information to be analyzed; Performing a convolution operation on the second abnormal performance feedback event thermal feature set to determine a dynamic performance mutation confidence weight of each second convolution kernel of the second abnormal performance feedback event thermal feature set; If the dynamic performance mutation confidence weight of the second convolution kernel is greater than the second set value, the second convolution feature corresponding to the dynamic performance mutation confidence weight of the second convolution kernel is used as the mask feature from the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed.

5. The servo controller fault data analysis method according to claim 3, characterized in that: The method of performing fault tracing on the operating environment sensor monitoring information to be analyzed by using the target operation and maintenance fault tracing model to obtain a mask feature of the fault tracing viewpoint of the servo controller fault event in the operating environment sensor monitoring information to be analyzed includes: Through the target operation and maintenance fault tracing model, the fault tracing is performed on the operating environment sensor monitoring information to be analyzed to obtain a second abnormal performance feedback event thermal feature set of the operating environment sensor monitoring information to be analyzed; According to the thermal feature set of the second abnormal performance feedback event, the convolution feature with the largest abnormal performance feedback thermal coefficient is obtained as a mask feature from the fault tracing perspective of the servo controller fault event in the operating environment sensor monitoring information to be analyzed.

6. The servo controller fault data analysis method according to claim 1, characterized in that: The priori environment monitoring log includes first priori operating environment sensor monitoring information and second priori operating environment sensor monitoring information. According to the servo controller fault event category characteristics, the fault performance variation involvement vector of the priori operating environment sensor monitoring information and the closed climate operating environment description semantics of the past operating environment sensor monitoring information are correlated and analyzed to obtain the environmental factor impact evaluation of the past operating environment sensor monitoring information, including: Obtaining the fault performance variation involvement vector of the first priori operating environment sensor monitoring information according to the servo controller fault event category characteristics of the first priori operating environment sensor monitoring information; obtaining the fault performance variation involvement vector of the second priori operating environment sensor monitoring information according to the servo controller fault event category characteristics of the second priori operating environment sensor monitoring information; Performing semantic embedding according to the past operating environment sensor monitoring information to obtain the closed climate operating environment description semantics of the past operating environment sensor monitoring information; Performing correlation analysis on the fault performance variation involvement vector of the first priori operating environment sensor monitoring information and the closed climate operating environment description semantics to obtain a first causal deduction weight between the first priori operating environment sensor monitoring information and the past operating environment sensor monitoring information; Performing correlation analysis on the fault performance variation involvement vector of the second priori operating environment sensor monitoring information and the closed climate operating environment description semantics to obtain a second causal deduction weight between the second priori operating environment sensor monitoring information and the past operating environment sensor monitoring information; The first causal deduction weight and the second causal deduction weight are weighted to obtain the environmental factor impact evaluation.

7. The servo controller fault data analysis method according to claim 6, characterized in that: The servo controller fault event is a physical fault event, and the semantic embedding is performed based on the past operating environment sensor monitoring information to obtain the closed climate operating environment description semantics of the past operating environment sensor monitoring information, including: Performing physical environment sensor feature mining on the past operating environment sensor monitoring information to obtain a physical environment sensor feature set of the past operating environment sensor monitoring information; The physical environment sensing feature set is semantically embedded to obtain the closed climate operating environment description semantics of the past operating environment sensing monitoring information.

8. The servo controller fault data analysis method according to claim 1, characterized in that: The step of optimizing the initial operation and maintenance fault tracing model based on the first tracing discrimination thermal matrix and the past operating environment sensor monitoring information to obtain a target operation and maintenance fault tracing model includes: Through the initial operation and maintenance fault tracing model, the fault tracing is performed on the past operating environment sensor monitoring information to obtain a first tracing prediction thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information; Obtaining a first model training error of the initial operation and maintenance fault tracing model according to the first tracing prediction thermal matrix and the first tracing discrimination thermal matrix; Optimizing the initial operation and maintenance fault tracing model according to the first model training error to obtain a target operation and maintenance fault tracing model; The initial operation and maintenance fault tracing model is used to trace the fault source of the past operating environment sensor monitoring information to obtain a first tracing prediction thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information, including: The past operating environment sensor monitoring information is semantically embedded through the initial operation and maintenance fault tracing model to obtain the original environment monitoring vector of the past operating environment sensor monitoring information; Through the initial operation and maintenance fault tracing model, the original environment monitoring vector of the past operating environment sensor monitoring information is fully connected to obtain the environment monitoring fully connected vector of the past operating environment sensor monitoring information; Perform feature upsampling based on the original environment monitoring vector and the environment monitoring fully connected vector to obtain a first traceability prediction thermal matrix of the past operating environment sensor monitoring information; The optimizing the initial operation and maintenance fault tracing model according to the first model training error to obtain a target operation and maintenance fault tracing model includes: According to the servo controller fault event category characteristics of the prior operating environment sensor monitoring information, a second tracing discrimination thermal matrix of the fault event triggering feature set in the prior operating environment sensor monitoring information is obtained; Through the initial operation and maintenance fault tracing model, the prior operating environment sensor monitoring information is traced to the fault source, and a second tracing prediction thermal matrix of the fault event triggering feature set in the prior operating environment sensor monitoring information is obtained; Obtaining a second model training error of the initial operation and maintenance fault tracing model according to the second tracing prediction thermal matrix and the second tracing discrimination thermal matrix; According to the first model training error and the second model training error, the initial operation and maintenance fault tracing model is optimized to obtain a target operation and maintenance fault tracing model.

9. The servo controller fault data analysis method according to any one of claims 1 to 8, characterized in that: The first tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information is obtained based on the fault tracing viewpoint thermal network, including: According to the thermal network of the fault tracing point of view, the past operating environment sensor monitoring information is quantized to obtain a quantized monitoring vector set of the past operating environment sensor monitoring information; The quantized monitoring vector set is subjected to a thermal space migration transformation to obtain a first source tracing discrimination thermal matrix of the fault event triggering feature set in the past operating environment sensor monitoring information.

10. An environmental sensing analysis system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 9.

Images