A method, system, electronic device, and storage medium for predicting industrial equipment failures.

Abstract

This application relates to the field of industrial equipment failure prediction, and more specifically, to a method, system, electronic device, and storage medium for industrial equipment failure prediction. The method includes: acquiring operational data, environmental data, historical failure data, and spare parts data of the industrial equipment; extracting features from the operational data, historical failure data, and spare parts data to obtain operational features, historical failure correlation features, and spare parts correlation features; obtaining operational correction features based on a dual model of operational features, environmental data, and environmental interference processing; inputting the operational correction features into a long short-term memory network to obtain a first prediction result; inputting the operational correction features, the first prediction result, historical failure correlation features, and spare parts correlation features into a gradient booster model to obtain a second prediction result; and inputting the first prediction result and the second prediction result into a large industrial model to obtain a third prediction result. This method can achieve accurate prediction of industrial equipment failures based on the processing of multi-dimensional data.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment fault prediction, and more specifically, to a method, system, electronic device, and storage medium for industrial equipment fault prediction. Background Technology

[0002] In modern manufacturing, discrete manufacturing is a production model that assembles parts into final products through multiple processes. Its key characteristics are discrete production processes and complex product structures, relying on highly automated industrial equipment clusters for efficient operation. Typical industrial equipment includes CNC machine tools for precision machining, robotic arms for material handling and assembly, and Automated Guided Vehicles (AGVs) for intelligent logistics scheduling. With the deepening of intelligent manufacturing transformation, the reliability of industrial equipment has become a key factor restricting production efficiency, especially for industrial equipment in discrete manufacturing scenarios, where fault prediction faces multiple technical challenges. Due to the significant heterogeneity of equipment operating data—including physical parameters such as time-series vibration signals and current fluctuations, as well as process parameter adjustment records—and the fact that early fault characteristics are often hidden in normal operating noise, traditional predictive maintenance solutions are often limited by single-dimensional data acquisition methods, only able to capture local anomalies and struggling to obtain the full lifecycle health status of industrial equipment. Furthermore, existing technologies generally analyze the physical state of industrial equipment in isolation, failing to effectively utilize supply chain data such as spare parts inventory turnover and maintenance work order response time, leading to discrepancies between prediction results and actual conditions. This fragmented data across various dimensions not only leads to frequent unplanned shutdowns and production line disruptions, but also results in redundant inventory or shortages due to inaccurate spare parts reserve strategies. Under this dual pressure, the contradiction between rising operating costs and extended delivery cycles is exacerbated. Summary of the Invention

[0003] This invention provides an industrial equipment fault prediction method, system, electronic device, and storage medium, which are used to achieve accurate prediction of industrial equipment faults based on the processing of multidimensional data.

[0004] According to a first aspect of this application, an industrial equipment failure prediction method is provided, the method comprising:

[0005] Acquire operational and environmental data of the industrial equipment, and acquire historical fault data and spare parts data related to the industrial equipment.

[0006] The operating data is subjected to feature extraction to obtain operating features; the historical fault data is subjected to feature extraction to obtain historical fault association features; and the spare parts data is subjected to feature extraction to obtain spare parts association features.

[0007] The operational correction features are obtained based on the operational characteristics, the environmental data, and the preset dual model for handling environmental interference.

[0008] The operation correction features are input into a preset long short-term memory network to obtain a first prediction result;

[0009] The operation correction features, the first prediction result, the historical fault association features, and the spare parts association features are input into a preset gradient booster model to obtain a second prediction result;

[0010] The first prediction result and the second prediction result are input into a preset industrial large model to obtain a third prediction result.

[0011] Understandably, by acquiring operational data, environmental data, historical fault data, and spare parts data of industrial equipment, a multi-dimensional feature system is constructed. This system, coupled with a dual-model approach to dynamically correct operational features using environmental interference processing, effectively filters out external noise such as temperature and humidity fluctuations and electromagnetic interference, significantly improving the robustness and reliability of operational features. Based on this, a hierarchical construction of a long short-term memory network, a gradient booster model, and a large industrial model is established, forming a closed-loop prediction process for fault prediction and root cause identification of industrial equipment. This enables high-accuracy fault warnings within a rapid response time.

[0012] Optionally, before the step of extracting features from the running data to obtain running features, the method further includes:

[0013] Obtain the location points corresponding to the missing data in the running data;

[0014] When the number of points at the locations of several consecutive missing data is less than or equal to a preset number threshold, the missing data is filled in using interpolation to obtain the filled running data, and outliers in the filled running data are removed based on preset rules to obtain the preprocessed running data.

[0015] When the number of points with consecutive missing data exceeds the threshold, the running data is discarded, and the missing data situation is reported.

[0016] The step of extracting features from the running data to obtain running features specifically involves extracting features from the preprocessed running data to obtain running features.

[0017] Understandably, a data preprocessing mechanism was designed before feature extraction. By dynamically identifying continuous missing data in the running data, and combining it with preset point thresholds, hierarchical processing was flexibly implemented: for small-scale missing data, interpolation was used to accurately complete the missing data and automatically remove outliers to ensure the continuity and purity of the running feature input; when the missing data scale exceeded the standard, inferior data was actively discarded and real-time alarms were triggered, effectively blocking the transmission of low-quality information to the prediction stage, significantly improving the reliability of the running data, avoiding the risk of misjudgment caused by data incompleteness or noise interference in traditional methods, providing a high-fidelity feature foundation for subsequent fault prediction, and ultimately improving the stability of prediction and the accuracy of decision-making.

[0018] Optionally, the dual model for environmental interference processing includes a linear environmental interference compensation model and a nonlinear environmental interference separation model;

[0019] The step of obtaining operational correction features based on the operational characteristics, the environmental data, and a preset dual-model of environmental interference includes:

[0020] The environmental data is input into the linear environmental disturbance compensation model to obtain the correction amount, and the operating characteristics are corrected according to the correction amount to obtain the first intermediate correction feature;

[0021] The environmental data and the operational characteristics are input into the nonlinear environmental disturbance separation model to obtain the second intermediate correction feature;

[0022] The first intermediate correction feature and the second intermediate correction feature are weighted and fused according to preset weights to obtain the running correction feature.

[0023] Understandably, the dual-mode collaborative architecture of constructing a linear environmental disturbance compensation model and a nonlinear environmental disturbance separation model achieves accurate separation and compensation of environmental disturbances by dynamically analyzing the impact of external noise such as temperature and humidity fluctuations and electromagnetic interference on operational characteristics. The linear environmental disturbance compensation model corrects quantifiable environmental offsets in real time, while the nonlinear environmental disturbance separation model deeply mines hidden interference factors in complex operating conditions. The linear environmental disturbance compensation model and the nonlinear environmental disturbance separation model are also weighted and fused with preset weights to generate high-fidelity operational correction features, which significantly enhances the robustness and anti-interference ability of operational features. This effectively solves the problem of early fault missed detection or false alarm caused by single-dimensional data collection in traditional methods, laying a highly reliable data foundation for subsequent fault prediction.

[0024] Optionally, the method further includes:

[0025] Several fault severity levels and several fault types are preset;

[0026] The first prediction result includes the predicted occurrence time of the industrial equipment failure and the severity level of the failure matched by the industrial equipment; the second prediction result includes the failure type and failure location matched by the industrial equipment; and the third prediction result includes the failure diagnosis, failure handling plan and spare parts allocation plan for the industrial equipment.

[0027] Understandably, by pre-setting multiple levels of fault severity and diverse fault types, a structured diagnostic framework is constructed. This enables the prediction results to not only cover the fault occurrence time and severity level, but also accurately locate the fault type, specific location, fault diagnosis, fault handling plan, and spare parts allocation plan. This achieves full-dimensional coverage from early warning to operation and maintenance processing analysis, thereby significantly improving the granularity and interpretability of fault diagnosis, avoiding the risk of misjudgment caused by fuzzy classification in traditional methods, and providing a clear basis for maintenance decisions.

[0028] Optionally, the method further includes:

[0029] Based on a preset algorithm, feature filtering is performed on the operation correction features, the historical fault association features, and the spare parts association features to obtain the operation core features, the historical fault association core features, and the spare parts association core features.

[0030] The step of inputting the operational correction features into a preset long short-term memory network to obtain a first prediction result specifically involves: inputting the operational core features into a preset long short-term memory network to obtain a first prediction result;

[0031] The second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into a preset gradient booster model. Specifically, the second prediction result is obtained by inputting the operation core feature, the first prediction result, the historical fault association core feature, and the spare parts association core feature into a preset gradient booster model.

[0032] Understandably, by using a pre-defined algorithm, core features are accurately extracted from operational correction features, historical fault association features, and spare parts association features. This effectively preserves core features that are strongly correlated with the fault types of industrial equipment, significantly improves the purity and representativeness of feature data, and significantly optimizes the accuracy of capturing the operational degradation patterns of industrial equipment and classifying fault types. At the same time, it reduces the consumption of computing resources and accelerates the prediction response, realizes efficient collaboration of multi-source heterogeneous data, and provides a more reliable and real-time decision-making basis for industrial equipment fault prediction.

[0033] Optionally, the method further includes:

[0034] Acquire maintenance knowledge data corresponding to industrial equipment, and convert the maintenance knowledge data into several adjustment instructions;

[0035] Obtain adjustment instructions that match the aforementioned operational correction features;

[0036] Based on the matching adjustment instruction, the prediction weights corresponding to the adjustment instruction in the Long Short-Term Memory Network and the Gradient Boosting Machine Model are adjusted to obtain the adjusted Long Short-Term Memory Network and the adjusted Gradient Boosting Machine Model.

[0037] The operation correction features are input into a preset long short-term memory network to obtain a first prediction result, specifically: the operation correction features are input into an adjusted long short-term memory network to obtain a first prediction result;

[0038] The second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into a preset gradient booster model. Specifically, the second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into an adjusted gradient booster model.

[0039] Understandably, by transforming maintenance knowledge data into dynamic adjustment instructions and matching and correcting features in real time, adaptive calibration of the prediction weights of the Long Short-Term Memory Network (LSTM) and Gradient Boosting Machine (GBR) models is achieved. This enables the LSTM and GBR models to accurately respond to the dynamic changes in the operating conditions of industrial equipment, thereby significantly improving the sensitivity and classification accuracy of fault trend capture. It effectively eliminates the false alarms or missed detections caused by parameter fixation in traditional models, while also greatly shortening the prediction response delay.

[0040] Optionally, after the step of inputting the first prediction result and the second prediction result into a preset industrial large-scale model to obtain the third prediction result, the method further includes:

[0041] A fault report corresponding to the industrial equipment is generated based on the first prediction result and / or the second prediction result and / or the third prediction result.

[0042] Understandably, by deeply integrating the multi-layered prediction results of long short-term memory networks, gradient booster models, and large industrial models, a comprehensive report covering failure time, type, diagnosis, maintenance, and spare parts allocation can be intelligently generated. This enables fully automated decision-making from early warning to execution, significantly reducing manual intervention and improving response time. Consequently, it will reduce unplanned downtime, simultaneously optimize spare parts inventory turnover efficiency, and provide a precise and low-cost industrial equipment health management system for discrete manufacturing scenarios.

[0043] According to a second aspect of this application, an industrial equipment fault prediction system is provided, the system comprising:

[0044] The data acquisition module is used to acquire the operating data and environmental data of the industrial equipment, as well as the historical fault data and spare parts data related to the industrial equipment.

[0045] The feature extraction module is used to extract features from the operating data to obtain operating features; to extract features from the historical fault data to obtain historical fault association features; and to extract features from the spare parts data to obtain spare parts association features.

[0046] The correction module is used to obtain operational correction features based on the operational features, the environmental data, and a preset dual model for handling environmental interference.

[0047] The prediction module is used to input the operation correction features into a preset long short-term memory network to obtain a first prediction result; input the operation correction features, the first prediction result, the historical fault association features, and the spare parts association features into a preset gradient booster model to obtain a second prediction result; and input the first prediction result and the second prediction result into a preset industrial large model to obtain a third prediction result.

[0048] According to a third aspect of this application, an electronic device is provided, comprising:

[0049] Memory, used to store one or more computer programs;

[0050] A processor, when the one or more computer programs are executed by the processor, implements the industrial equipment fault prediction method described in the first aspect above.

[0051] According to a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the industrial equipment fault prediction method described in the first aspect above.

[0052] Based on any of the above aspects, the industrial equipment fault prediction method, system, electronic device, and storage medium provided in this application embodiment can achieve the following beneficial effects:

[0053] A dual-model approach for handling environmental interference is employed to analyze the operating characteristics of industrial equipment and environmental data, effectively eliminating the interference of external noise on feature extraction: the linear environmental interference compensation model accurately compensates for quantifiable environmental offsets, while the nonlinear environmental interference separation model separates hidden interference factors in complex operating conditions. After weighted fusion, high-fidelity operating correction features are generated, significantly improving the robustness and reliability of the operating correction features. This solves the problems of high false alarm rate and missed early fault detection caused by single-dimensional data collection in traditional methods, laying a precise data foundation for subsequent prediction.

[0054] A hierarchical architecture integrating Long Short-Term Memory (LSTM) networks, Gradient Boosting Machine (GPX) models, and a large industrial model achieves a balance between accuracy and real-time performance in predicting industrial equipment failures. The LSM network deeply mines dependencies in operational correction features to accurately predict failure occurrence time and severity. The GPX model quickly classifies industrial equipment failure types and locates the failures. The large industrial model coordinates all input prediction results, outputting industrial failure diagnosis, failure handling solutions, and spare parts allocation plans, ultimately generating a failure report. Furthermore, the LSM network, GPX model, and large industrial model work collaboratively through feature selection and weight adjustment mechanisms, responding quickly to and completing the entire chain of analysis—from failure prediction and failure type identification to failure handling decision generation—avoiding the accuracy limitations of single models and overcoming the latency bottlenecks of complex algorithms, ensuring both high accuracy and applicability to industrial scenarios.

[0055] Historical fault data and spare parts data are deeply correlated with fault prediction, forming a predictive foundation for equipment failures and faulty spare parts: historical fault data reveals the degradation patterns of industrial equipment, while spare parts data reflects the real-time status of available resources. After feature filtering, both are input into the prediction model, which can not only provide early warnings of potential faults but also automatically generate the optimal fault handling solution matching the current inventory level. This completely reverses the drawbacks of the separation between prediction and warehouse management in traditional maintenance, thereby reducing unplanned downtime, improving spare parts turnover efficiency, simplifying manual decision-making processes, transforming complex maintenance tasks into standardized operating instructions, and significantly reducing operation and maintenance costs and response cycles. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of an industrial equipment fault prediction method provided in this embodiment.

[0058] Figure 2 This is a flowchart for obtaining runtime correction features provided in this embodiment.

[0059] Figure 3 The flowchart shows the adjustment process for the Long Short-Term Memory network and Gradient Boosting Machine model provided in this embodiment.

[0060] Figure 4 This is a schematic diagram of the functional modules of an industrial equipment fault prediction system provided in this embodiment.

[0061] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this embodiment. Detailed Implementation

[0062] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate the following embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0063] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0064] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0065] In the discrete manufacturing scenario of modern manufacturing, failure prediction of industrial equipment faces significant challenges. It is necessary to overcome the heterogeneity of operating data and the concealment of early failure characteristics. At the same time, traditional solutions fail to effectively integrate supply chain data due to the isolated analysis of the status of industrial equipment, resulting in prediction bias, which in turn leads to problems such as unplanned downtime and inaccurate inventory, exacerbating the contradiction between operating costs and delivery cycle.

[0066] This embodiment provides a technical solution that can solve the above problems. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.

[0067] like Figure 1 As shown, this embodiment provides a method for predicting industrial equipment failures, which can be further divided into the following steps:

[0068] S100: Obtain the operating data and environmental data of the industrial equipment, and obtain historical fault data and spare parts data related to the industrial equipment.

[0069] For example, in this embodiment, the industrial equipment may include CNC machine tools, robotic arms, and guided transport vehicles. Acquiring operational and environmental data from the industrial equipment requires data collection from sensors installed on or near the equipment.

[0070] Based on historical experience, the most frequent failure points of CNC machine tools are spindle bearings, feed axis guideways, and servo motors. Therefore, sensors can be installed at these high-failure points to collect operational and environmental data. For example, the sensors collecting operational data can be vibration sensors and current sensors; a vibration sensor (PCB 356A16) and a current sensor (LEM LA 100-P) can be used. The vibration sensor can be magnetically fixed to the spindle end cover of the CNC machine tool, collecting vibration data at a preset first sampling frequency, which can be set to 500Hz and adjusted as needed. The current sensor can be embedded through a preset interface on the CNC machine tool, collecting current data at a preset second sampling frequency, which can be set to 10Hz and adjusted as needed. For example, the sensor collecting environmental data can include a temperature sensor; a PT1000 armored temperature sensor can be used. The temperature sensor can be embedded in the CNC machine tool via a pre-set interface and collects temperature environmental data at the third sampling frequency, which can be set to 10Hz and adjusted appropriately according to actual conditions. Based on historical experience, the CNC machine tool can operate normally in environments ranging from 0 to 150 degrees Celsius, is resistant to cutting fluid corrosion, meets IP67 protection rating, and is resistant to electromagnetic interference, meeting standard IEC 61000-6-2. The above-mentioned normal operating conditions of the CNC machine tool can provide reference data for subsequent fault prediction, thereby improving the accuracy of fault prediction.

[0071] In this embodiment, the robotic arm can be an assembly line robotic arm. Based on historical experience, common failure points of assembly line robotic arms include joint bearings, transmission gearboxes, and end effectors. Failures in these areas can cause the robotic arm to stop operating. Therefore, sensors can be installed at these common failure points to collect operational data. For example, the sensors collecting operational data can include acoustic sensors, vibration sensors, and torque sensors. Acoustic sensors (model B&K 4944), vibration sensors (model IEPE accelerometer), and torque sensors (model HBM T40B) can be used. These sensors can be fixed to non-moving parts of the assembly line robotic arm using dedicated brackets, at a certain distance from the common failure points. This distance can be less than or equal to 3 cm and can be adjusted appropriately based on actual conditions. Furthermore, a fourth sampling frequency is preset to collect acoustic operation data, selected within the range of 200Hz to 10kHz; a fifth sampling frequency is preset to collect vibration operation data, which can be set to 500Hz and adjusted appropriately according to actual conditions; a sixth sampling frequency is preset to collect torque operation data, which can be set to 10Hz and adjusted appropriately according to actual conditions. Based on historical experience, the assembly workshop robotic arm can withstand impacts of up to 10 times the force of gravity during normal operation, meeting the IP65 protection rating and adapting to the dusty environment of the workshop. The above-mentioned conditions for the normal operation of the assembly workshop robotic arm can provide reference data for subsequent fault prediction, thereby improving the accuracy of fault prediction.

[0072] In this embodiment, the guided vehicle can be a warehouse guided vehicle (AGV). Based on historical experience, the most common malfunctions in AGVs are found in the drive wheel motor, steering joint, and forklift mechanism. Therefore, sensors can be installed at these malfunction-prone areas to collect environmental and operational data. For example, the sensors used to collect operational and environmental data may include a temperature and vibration integrated sensor, a speed sensor, and a current sensor. A Schneider Electric A9MEM3100 temperature and vibration integrated sensor, a photoelectric E3Z-LS63 speed sensor, and a LEM LV25-P current sensor can be used. The temperature and vibration integrated sensor can be bolted to the drive motor housing, and the speed sensor and current sensor can be integrated into key parts of the AGV chassis. Furthermore, a seventh sampling frequency is preset to collect temperature and vibration data. This seventh sampling frequency can be 200Hz and can be adjusted appropriately based on actual conditions. A preset eighth sampling frequency is used to collect rotational speed and current data. This eighth sampling frequency can be set to 5Hz and can be adjusted appropriately based on actual conditions. Based on historical experience, the robotic arm in the assembly workshop can withstand ambient temperatures ranging from -10°C to 60°C during normal operation and can withstand vibrations at five times the force of gravity, meeting the IP66 protection rating. The above-mentioned conditions for the normal operation of the warehouse guided transport vehicle can provide reference data for subsequent fault prediction, thereby improving the accuracy of fault prediction.

[0073] It is understood that, based on the above description, the operational data may include the operational data of CNC machine tools, the operational data of robotic arms in assembly workshops, and the operational data of warehouse guided transport vehicles. In order to ensure timely response after industrial equipment malfunctions, in practical applications, real-time operational data of industrial equipment is collected to predict malfunctions with the fastest possible response speed, minimizing the impact of industrial equipment failures on manufacturing production.

[0074] In practical implementation, a Manufacturing Execution System (MES) is typically set up to monitor the status and tasks of all industrial equipment in real time, intelligently coordinate production scheduling and maintenance response, and seamlessly link processes such as issuing early warnings, generating maintenance work orders, and supplying spare parts based on operational data, thus facilitating a rapid closed loop from prediction to maintenance. In practical applications, historical fault data related to the industrial equipment can be obtained from the MES.

[0075] In practical implementation, a Warehouse Management System (WMS) is typically set up to achieve precise warehouse location management and automated scheduling, ensuring that raw materials, work-in-process, and finished products flow at the right time with maximum efficiency. This enables real-time visibility of spare parts inventory and unmanned collaborative processes, providing seamless spare parts supply assurance. In actual applications, spare parts data related to the industrial equipment can be obtained from the Warehouse Management System.

[0076] Preferably, to adapt to the layout logic of various industrial devices in discrete manufacturing scenarios, considering the large number of industrial devices, the need to collect and upload a large amount of data, and the diverse distribution of these devices, including potential movement of some equipment such as warehouse guided transport vehicles, this embodiment also utilizes a data acquisition gateway to collect sensor data. The data acquisition gateway can aggregate the data collected by each sensor and perform protocol conversion. For example, the gateway can support common discrete manufacturing protocols such as Modbus-RTU and Profinet IRT; it also interfaces with the equipment management center and the warehouse management center to achieve bidirectional synchronization of operational data and spare parts data. The warehouse management center can be one or more databases, such as MySQL, used to store maintenance information such as spare parts data for industrial equipment.

[0077] S200: Extract features from the operating data to obtain operating features; extract features from the historical fault data to obtain historical fault association features; extract features from the spare parts data to obtain spare parts association features;

[0078] Understandably, considering the real-time nature of industrial equipment fault prediction and the massive workload of uploading sensor data from a large number of industrial devices, in this embodiment, edge computing nodes can be set up to perform preliminary data processing. The data that has undergone preliminary processing is then transmitted to the cloud for fault prediction, thereby ensuring rapid data transmission and efficient fault identification.

[0079] Understandably, the operational data is collected from sensors on industrial equipment, transmitted to a data acquisition gateway, and then uploaded to an edge computing node for feature extraction to obtain operational features. Similarly, the environmental data is collected from sensors on industrial equipment, transmitted to a data acquisition gateway, and then uploaded to an edge computing node to correct the operational features.

[0080] In this embodiment, since historical fault data is obtained through the connection to the equipment management center and does not require connection to on-site sensors, it can be obtained from the cloud by connecting to the equipment management center to participate in subsequent fault prediction work. Similarly, spare parts data is obtained through the connection to the warehouse management center and does not require connection to on-site sensors, so it can be obtained from the cloud by connecting to the warehouse management center to participate in subsequent fault prediction work.

[0081] In this embodiment, feature extraction of the operational data to obtain operational features can be performed at an edge computing node, while feature extraction of the historical fault data to obtain historical fault association features and feature extraction of the spare parts data to obtain spare parts association features can be performed in the cloud.

[0082] Specifically, before the step of extracting features from the operational data to obtain operational features, the method includes:

[0083] Obtain the location points corresponding to the missing data in the running data;

[0084] When the number of points at the locations of several consecutive missing data is less than or equal to a preset number threshold, the missing data is filled in using interpolation to obtain the filled running data, and outliers in the filled running data are removed based on preset rules to obtain the preprocessed running data.

[0085] When the number of points with consecutive missing data exceeds the threshold, the running data is discarded, and the missing data situation is reported.

[0086] In step S200, feature extraction is performed on the running data to obtain running features. Specifically, feature extraction is performed on the preprocessed running data to obtain running features.

[0087] It is understood that the preprocessing work can all be completed at the edge computing nodes. In this embodiment, the sensor collects operational data based on a preset acquisition frequency. The data collected at each time point is considered as data for one location point, and the data corresponding to the preset number of data points are integrated into one operational data. That is, the operational data contains data corresponding to multiple location points. If there are network problems or sensor malfunctions, data may not be collected, resulting in missing data for some location points. The location points corresponding to the missing data in the operational data are obtained. When the number of location points with several consecutive missing data is less than or equal to a preset number threshold, it indicates that the network is experiencing a brief transmission problem or the sensor is experiencing a brief malfunction, and data transmission and acquisition have resumed at other times. Therefore, the missing data can be filled in using interpolation to obtain the completed operational data. Preferably, the interpolation method can be the adjacent time-time parameter interpolation method. The number threshold can be preset to be less than or equal to 5 location points, and can be adjusted appropriately according to the actual situation.

[0088] In this embodiment, it is also necessary to remove outliers from the completed running data based on preset rules to obtain preprocessed running data. This is because momentary malfunctions may occur during sensor acquisition, resulting in abrupt changes in the acquired data. These abrupt changes, where the difference from the preceding and following data exceeds a preset threshold and the duration is relatively short, are considered acquisition error data and need to be removed.

[0089] In this embodiment, when the number of locations with consecutive missing data exceeds the threshold, it indicates that the sensor has been malfunctioning for an extended period, or the corresponding network connection has been faulty for a long time. Therefore, the operational data obtained by the edge computing node cannot carry information and needs to be discarded to avoid affecting subsequent fault prediction. Simultaneously, the edge computing node needs to report the missing operational data, specifically to the equipment management center. The equipment management center issues an alarm based on the missing data, queries the operating status of the industrial equipment corresponding to the missing data, and generates a maintenance work order to remind maintenance personnel to locate and repair the equipment.

[0090] S300. Based on the operational characteristics, the environmental data, and the preset dual model for handling environmental interference, operational correction characteristics are obtained;

[0091] In this embodiment, in order to eliminate the false interference of environmental factors on the operating characteristics of industrial equipment, it is necessary to construct a dual model for environmental interference processing to achieve accurate purification of operating characteristics and improve the reliability of operating characteristics.

[0092] Specifically, the dual model for environmental interference processing includes a linear environmental interference compensation model and a nonlinear environmental interference separation model;

[0093] Understandably, before practical application, the dual model for handling environmental interference needs to be trained and tested to obtain a well-trained dual model for handling environmental interference, thereby improving the efficiency of correcting operational features during practical application. Preferably, a first training set needs to be pre-constructed. Each training sample in the training set contains an environmental data training vector, an operational feature training vector, and a fault label. The environmental data training vector can be extracted from historical environmental data. For example, the historical environmental data can include workshop ambient temperature, humidity, dust concentration, power grid voltage fluctuation amplitude, vibration interference intensity of nearby industrial equipment, cutting fluid protection status, ground flatness level, duration of environmental interference, etc. The environmental data training vector is extracted using this environmental data. The operational feature training vector can be extracted from historical operational features. The fault label is pre-set, and the fault level of the training sample is marked according to the actual situation of the industrial equipment. The fault level can include normal level, minor fault level, and severe fault level. The fault level is used as the fault label to label the training sample.

[0094] Meanwhile, the training samples are expanded through environmental scenario expansion strategies. For example, environmental interference of different intensities is superimposed on the collected operational data. For instance, operational data with temperature fluctuations of 5°C, power grid fluctuations of 3V, and vibration interference from nearby industrial equipment of 0.2g are added to simulate the operating state of industrial equipment in a complex workshop environment. The size of the first training set is expanded to a preset multiple of the original size to improve the generalization ability of the dual model for handling environmental interference.

[0095] Understandably, the linear environmental interference compensation model primarily compensates for linear interferences such as temperature and power grid fluctuations in operational characteristics. Preferably, the linear environmental interference compensation model can employ a gradient boosting machine (XGBoost) regression model. By learning the linear correlation between environmental data and operational characteristics, it establishes a mapping relationship between environmental changes in the environmental data training vector and the offset of the operational characteristic training vector. This allows for the correction of the operational characteristic training vector to obtain a first intermediate corrected training feature. For example, this mapping relationship could be that for every 1°C increase in ambient temperature, the linear offset of the motor winding temperature is 0.8°C. In subsequent inference stages, the input environmental data is identified, and the corresponding correction amount is used to correct the operational characteristics, thereby obtaining the first intermediate corrected feature.

[0096] Understandably, the nonlinear environmental interference separation model is mainly used to separate nonlinear interferences such as vibration and humidity from nearby industrial equipment. Preferably, the nonlinear environmental interference separation model can employ a lightweight convolutional neural network (CNN) based on an attention mechanism. This network's attention layer automatically learns the correlation between environmental data training vectors and operational feature training vectors. Based on this correlation, it obtains interference features in the operational feature training vectors that are affected by environmental data training vectors related to nonlinear interference, reducing the weight of these interference features in the operational feature training vectors to suppress interference. Simultaneously, it learns and obtains key features in the operational feature training vectors that are highly correlated with industrial equipment faults, increasing the weight of these key features in the operational features to improve the accuracy of subsequent fault prediction. Thus, in the subsequent inference stage, the input environmental data and operational features are processed, and the operational features are corrected to obtain a second intermediate corrected feature.

[0097] Understandably, the trained dual-model for environmental interference is weighted, deployed to edge computing nodes, and then the running features are corrected after they are obtained at the edge computing nodes, thereby improving the reliability of the running features and the efficiency of processing them.

[0098] Preferably, an iterative mechanism can also be set for the dual environmental interference model, that is, environmental data, running features - running correction features - first prediction results and second prediction results are periodically uploaded to the cloud in the edge computing node, and the dual environmental interference model is incrementally trained based on the newly added data at preset time intervals to update the model parameters, so as to ensure that the compensation accuracy and separation accuracy of the dual environmental interference model are continuously optimized.

[0099] like Figure 2 As shown, obtaining the operational correction features based on the operational characteristics, the environmental data, and the preset dual model of environmental interference may include the following steps:

[0100] S310. Input the environmental data into the linear environmental interference compensation model to obtain the correction amount, and correct the operating characteristics according to the correction amount to obtain the first intermediate correction feature;

[0101] Preferably, after obtaining the correction amount, the first intermediate correction feature can be the running feature minus the correction amount.

[0102] S320. Input the environmental data and the operational characteristics into the nonlinear environmental disturbance separation model to obtain the second intermediate correction feature;

[0103] Preferably, the environmental data and the operational features are concatenated to obtain input data with the sum of the environmental data and operational features dimensions. Based on the processing of the nonlinear environmental interference separation model, a second intermediate correction feature with the same output dimension as the operational dimension is output.

[0104] S330. The first intermediate correction feature and the second intermediate correction feature are weighted and fused according to a preset weight to obtain the running correction feature.

[0105] For example, the weights can be preset to 0.6 for the first intermediate correction feature obtained based on the linear environmental interference compensation model and 0.4 for the second intermediate correction feature obtained based on the nonlinear environmental interference separation model. The weighted first intermediate correction feature and the second intermediate correction feature are then fused to obtain the running correction feature, ensuring that different types of environmental interference can be effectively stripped away.

[0106] Specifically, predicting faults in industrial equipment requires the use of well-trained long short-term memory networks, gradient booster models, and large industrial models to obtain more accurate prediction results.

[0107] Preferably, for Long Short-Term Memory (LSTM) networks, to overcome the temporal correlation of industrial equipment failure development and the hidden nature of early operational features corresponding to failure characteristics, traditional LTM networks can be trained to focus on improving the ability to capture early failure trends in industrial equipment. Preferably, the LTM network is structured with a sequentially connected input layer, feature enhancement layer, LTM backbone network, fully connected layer, and output layer, using collected historical operational data as a sample set. Each sample in the sample set represents historical operational data over a preset time period, which covers the complete cycle of industrial equipment failure development. For example, the time period is 30 days, with data collected every 5 minutes (i.e., a time window of 720 seconds for a single training sample), and a sliding time window with a step size of 24 is used to obtain several training samples.

[0108] The sample set is fed into the input layer, and the input layer then transmits the input features to the feature enhancement layer.

[0109] Preferably, the feature enhancement layer can be constructed using 16 convolutional kernels of size 3 to extract local features from the input features, enhance the weak features of early faults, and thus filter out redundant signals from the normal operation of industrial equipment. For example, the weak features of early faults can be the slight drift of the dominant vibration frequency in the early stages of bearing wear.

[0110] Preferably, the long short-term memory backbone network is configured as a two-layer stacked structure, with 128 hidden units in the first layer and 64 hidden units in the second layer. Layer normalization is performed in the introduction layer to prevent gradient vanishing, and the dropout probability is set to 0.2. At the same time, a fault feature attention mechanism is added, which assigns 1.5 times the weight to the running features that are strongly correlated with historical faults, thereby obtaining intermediate data, and the intermediate data is input into the fully connected layer.

[0111] Preferably, the fully connected layer is configured as two layers, with 32 and 8 nodes in the two layers respectively, and the ReLU activation function is used to process the intermediate data to obtain the first prediction result of the training, which is then transmitted to the output layer.

[0112] The output layer outputs a first prediction result, which includes the predicted occurrence time of the industrial equipment failure and the severity level of the failure matched by the industrial equipment. The severity level of the failure matched by the industrial equipment is a preset fault label. In this embodiment, it can be the same as the preset fault level of the dual model for environmental interference processing described above, including normal level, minor fault level, and severe fault level.

[0113] Preferably, the Long Short-Term Memory (LSTM) network employs a loss function designed with mean squared error (MSE) and fault severity weights, assigning higher weights to prediction errors related to severe faults. For example, this weight is set to 2.5 to ensure the accuracy of early warning for severe faults. The AdamW optimizer is used, with an initial learning rate of 0.001. The LTM network is adjusted using a fault sample decay strategy, with 200 training epochs and a batch size of 32 to adapt to the distribution characteristics of fault training samples. The fault sample decay strategy can be as follows: after every 50 training epochs, the learning rate is reduced to 0.8 of its original value.

[0114] Preferably, pre-collected industrial equipment maintenance knowledge data can be input into the long short-term memory network, and the long short-term memory network can be optimized based on the industrial equipment maintenance knowledge data. The optimized long short-term memory network can be put into practical application to improve the accuracy of prediction.

[0115] S400, Input the operation correction features into a preset long short-term memory network to obtain a first prediction result;

[0116] In practical applications, the first prediction result is the same as the first prediction result of the training output mentioned above, that is, it includes the predicted occurrence time of the industrial equipment failure and the severity level of the failure matched by the industrial equipment failure.

[0117] To overcome the diverse types of faults and the hidden locations of faults in industrial equipment, the gradient booster model is trained to accurately predict the fault types and locations of industrial equipment.

[0118] Preferably, a sample set needs to be constructed for the gradient booster model. The data dimension of a single sample in the sample set is 20-dimensional, and includes the operating characteristics of industrial equipment, the first prediction result output by the long short-term memory network, the historical fault association features and spare parts association features extracted from the cloud.

[0119] Simultaneously, it is necessary to pre-define the mapping relationship between fault types and fault locations of industrial equipment based on collected historical fault data and expert experience. For example, the pre-define fault type for the first fault label is CNC machine tool bearing wear, and the corresponding fault location is the CNC machine tool spindle bearing. Preferably, a corresponding labeling rule is also set for each mapping relationship. For example, the labeling rule for the first fault label is that the vibration dominant frequency drift rate of the CNC machine tool is greater than 8%, and the root mean square value is greater than 1.5g, while the number of faults in this part of the CNC machine tool is more than 3 times.

[0120] Preferably, the core parameters of the gradient booster model are preset, for example:

[0121] The default parameters for the gradient booster machine model are:

[0122] Configure the gradient booster model to output multi-class probabilities: objective="multi:softprob"

[0123] Configure coverage for 12 high-frequency fault types: num_class=12

[0124] To prevent overfitting and adapt to 20-dimensional features: max_depth=6.

[0125] The learning rate is set to: learning_rate=0.05.

[0126] Set the number of base learners in the gradient booster model to: n_estimators=200.

[0127] The default parameters for fault classification adaptation of the gradient booster model are:

[0128] Set the sample sampling rate to retain more associated samples of fault types and operational characteristics: subsample=0.85;

[0129] Set the feature sampling rate to prioritize retaining strongly correlated features of faults: colsample_bytree=0.75;

[0130] Set a node splitting threshold to improve the gradient boosting machine model's sensitivity to identifying niche fault types (gamma=0.15);

[0131] Set category weights to assign higher weights to high-frequency severe fault types: scale_pos_weight=[1.2,1.2,1.5,1.5,1.3,1.3,1.4,1.4,1.1,1.1,1.6,1.6].

[0132] The sample set is input into a preset gradient booster model for processing to obtain a trained second prediction result. The second prediction result includes the fault type and fault location matched by the industrial equipment. The fault type and fault location are one or more of the preset mapping relationships between fault types and fault locations mentioned above.

[0133] Preferably, pre-collected industrial equipment maintenance knowledge data can be input into the gradient booster model, and the gradient booster model can be optimized based on the industrial equipment maintenance knowledge data. The optimized gradient booster model can then be put into practical application, thereby improving the accuracy of prediction.

[0134] S500: Input the operation correction feature, the first prediction result, the historical fault association feature and the spare parts association feature into the preset gradient booster model to obtain the second prediction result;

[0135] In practical applications, the second prediction result is the same as the second prediction result of the training output mentioned above, that is, it includes the fault type and fault location matched by the industrial equipment.

[0136] For large industrial models, in order to overcome the fact that using long short-term memory networks and gradient booster models can only obtain fault predictions but not decision information such as fault diagnosis, large industrial models are trained to obtain corresponding fault diagnosis.

[0137] For example, the industrial big data model can be implemented using existing industrial big data models, such as the Huawei Cloud Pangu Industrial Big Data Model, which is a dedicated model for L2-level equipment maintenance. It can pre-input unstructured data such as maintenance manuals, historical fault cases, and spare parts information related to the industrial equipment, and learn from the unstructured data to accurately output the causes and triggers of the faults.

[0138] Specifically, the industrial large-scale model needs to be trained based on a preset number of historical fault maintenance data entries to achieve fine-tuning. Each historical fault maintenance data entry can include "industrial equipment fault parameters + maintenance plan + spare parts allocation record + expert maintenance experience." For example, the historical fault maintenance data could be: "CNC machine tool spindle vibration peak value: 2.5g, fault type: bearing wear, maintenance plan: replace SKF 6312 bearing, spare parts allocation: retrieved from the workshop spare parts warehouse, dynamic balancing test required after replacement."

[0139] For example, when training and fine-tuning the industrial big model based on the historical fault maintenance data, an incremental fine-tuning mode is adopted, freezing 90% of the parameters of the bottom layer of the industrial big model, and only training the top layer of industrial equipment maintenance adaptation layer. The industrial equipment maintenance adaptation layer includes a fault-spare part matching submodule and a maintenance process generation submodule, and the optimization objective function is set as fault location accuracy, maintenance scheme feasibility and spare part allocation efficiency, so as to generate accurate fault diagnosis and fault handling schemes in practical applications.

[0140] Preferably, the industrial big model adopts a cloud-edge collaborative mode, that is, the lightweight computing part of the industrial big model is deployed on the edge computing node to reduce inference latency in practical applications and to output fault maintenance suggestions in real time; the complete industrial big model is deployed in the cloud to handle complex fault scenarios such as the diagnosis of multiple concurrent faults or rare faults.

[0141] S600. Input the first prediction result and the second prediction result into the preset industrial large model to obtain the third prediction result.

[0142] Specifically, the third prediction result includes fault diagnosis, fault handling plan, and spare parts allocation plan. In practical applications, it is also necessary to connect to the equipment management center and the warehouse management center in the cloud. The equipment management center obtains the preset handling plan corresponding to the fault of the industrial equipment, thereby generating the fault handling plan in the third prediction result; the warehouse management center obtains the spare parts data of the industrial equipment, thereby generating the spare parts allocation plan in the third prediction result.

[0143] Specifically, the method further includes:

[0144] Based on a preset algorithm, feature filtering is performed on the operation correction features, the historical fault association features, and the spare parts association features to obtain the operation core features, the historical fault association core features, and the spare parts association core features.

[0145] The step of inputting the operational correction features into a preset long short-term memory network to obtain a first prediction result specifically involves: inputting the operational core features into a preset long short-term memory network to obtain a first prediction result;

[0146] The second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into a preset gradient booster model. Specifically, the second prediction result is obtained by inputting the operation core feature, the first prediction result, the historical fault association core feature, and the spare parts association core feature into a preset gradient booster model.

[0147] In this embodiment, the preset algorithm can be a genetic algorithm, thereby enabling the selection of core operational features, core historical fault association features, and core spare parts association features from the operational correction features, the historical fault association features, and the spare parts association features. Preferably, the optimization objective focuses on fault prediction accuracy, that is, when features are encoded into binary vectors, features strongly correlated with industrial equipment fault types are preferentially retained. For example, the features strongly correlated with industrial equipment fault types include vibration dominant frequency drift rate and current harmonic distortion rate, etc.

[0148] Preferably, the fitness functions of the Long Short-Term Memory Network and the Gradient Boosting Machine Model can also be set. The fitness function is the mutual information value between the feature and the fault type and the contribution of the feature to the prediction accuracy. Through crossover and mutation operations, iterative optimization is carried out to finally select the core fault features, thereby improving the training efficiency of the Long Short-Term Memory Network and the Gradient Boosting Machine Model and improving the accuracy of identifying early faults in industrial equipment.

[0149] Specifically, such as Figure 3 As shown, the method further includes the following steps:

[0150] A1. Obtain maintenance knowledge data corresponding to industrial equipment, and convert the maintenance knowledge data into several adjustment instructions;

[0151] A2. Obtain adjustment instructions that match the aforementioned operational correction features;

[0152] A3. Based on the matching adjustment instruction, adjust the prediction weights in the Long Short-Term Memory Network and the Gradient Boosting Machine Model corresponding to the adjustment instruction to obtain the adjusted Long Short-Term Memory Network and the adjusted Gradient Boosting Machine Model;

[0153] In this embodiment, maintenance knowledge can be converted into several adjustment instructions using knowledge distillation technology. Typically, the maintenance knowledge is mapped to preset adjustment instructions, allowing the corresponding adjustment instructions to be obtained based on the maintenance knowledge. Further, the adjustment instructions are used to instruct the Long Short-Term Memory network and / or the Gradient Boosting Machine model to adjust the prediction weights. For example, the maintenance knowledge data could be: when the main vibration frequency of a CNC machine tool is 120Hz and the cumulative operating time exceeds 10,000 hours, bearing wear may occur. Converting this maintenance knowledge data into adjustment instructions could be: when the main vibration frequency of a CNC machine tool is greater than 120Hz and the cumulative operating time exceeds 10,000 hours, increase the prediction weight for the "bearing wear" fault type.

[0154] When the operational correction feature is obtained, it can be detected whether there is a matching adjustment instruction. If so, the prediction weights in the Long Short-Term Memory (LSTM) network and the Gradient Boosting Machine (GBooster) model corresponding to the adjustment instruction are adjusted based on the matching adjustment instruction, resulting in an adjusted LSM network and an adjusted GBooster model. For example, if the operational correction feature is "CNC machine tool vibration frequency 123Hz and cumulative operation 10001 hours", then the aforementioned adjustment instruction is matched. Based on the adjustment instruction, the prediction weight of the LSM network for "bearing wear" is increased to a preset multiple, and the judgment threshold of the fault label corresponding to "bearing wear" in the GBooster model is reduced by a preset percentage. Preferably, the preset multiple can be set to 1.8 times, and the preset percentage can be set to 10%. These can be adjusted appropriately according to actual conditions to enhance the accuracy of fault prediction.

[0155] The operation correction features are input into a preset long short-term memory network to obtain a first prediction result, specifically: the operation correction features are input into an adjusted long short-term memory network to obtain a first prediction result;

[0156] The second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into a preset gradient booster model. Specifically, the second prediction result is obtained by inputting the operation correction feature, the first prediction result, the historical fault association feature, and the spare parts association feature into an adjusted gradient booster model.

[0157] Specifically, after the step of inputting the first prediction result and the second prediction result into a preset industrial large-scale model to obtain the third prediction result, the method further includes:

[0158] A fault report corresponding to the industrial equipment is generated based on the first prediction result and / or the second prediction result and / or the third prediction result.

[0159] In this embodiment, the first prediction result and / or the second prediction result and / or the third prediction result can be combined in pairs or all three to form a fault report. Preferably, the fault report also needs to be uploaded to the equipment management center, whereby the equipment management center generates a maintenance work order based on the fault report, alerting maintenance personnel and enabling them to respond quickly and repair the fault.

[0160] like Figure 4 As shown in the illustration, this application also provides an industrial equipment fault prediction system. Optionally, the system includes:

[0161] The module comprises a data acquisition module 711, a feature extraction module 712, a correction module 713, and a prediction module 714, wherein:

[0162] The data acquisition module 711 is used to acquire the operating data and environmental data of the industrial equipment, and to acquire historical fault data and spare parts data related to the industrial equipment.

[0163] In this embodiment, the data acquisition module 711 can be used to perform... Figure 1 For a detailed description of the data acquisition module 711, please refer to the description of step S100 shown.

[0164] The feature extraction module 712 is used to extract features from the operating data to obtain operating features; to extract features from the historical fault data to obtain historical fault association features; and to extract features from the spare parts data to obtain spare parts association features.

[0165] In this embodiment, the feature extraction module 712 can be used to perform... Figure 1 For a detailed description of the feature extraction module 712, please refer to the description of step S200 shown.

[0166] The correction module 713 is used to obtain the operation correction features based on the operation features, the environmental data, and the preset dual model for environmental interference processing;

[0167] In this embodiment, the correction module 713 can be used to perform... Figure 1 For a detailed description of step S300, the correction module 713 can be found in the description of step S300.

[0168] Specifically, the correction module 713 is further configured to input the environmental data into the linear environmental interference compensation model to obtain a correction amount, correct the operating characteristics according to the correction amount to obtain a first intermediate correction feature; input the environmental data and the operating characteristics into the nonlinear environmental interference separation model to obtain a second intermediate correction feature; and perform weighted fusion of the first intermediate correction feature and the second intermediate correction feature according to a preset weight to obtain an operating correction feature.

[0169] In this embodiment, the correction module 713 can also be used to perform... Figure 2 For a detailed description of the correction module 713, please refer to the description of steps S310-S330 shown.

[0170] The prediction module 714 is used to input the operation correction features into a preset long short-term memory network to obtain a first prediction result; input the operation correction features, the first prediction result, the historical fault association features, and the spare parts association features into a preset gradient booster model to obtain a second prediction result; and input the first prediction result and the second prediction result into a preset industrial large model to obtain a third prediction result.

[0171] In this embodiment, the prediction module 714 can be used to perform... Figure 1 For a detailed description of the prediction module 714, see steps S400-S600 shown.

[0172] Specifically, the prediction module 714 can also be used to acquire maintenance knowledge data corresponding to industrial equipment and convert the maintenance knowledge data into several adjustment instructions; acquire adjustment instructions that match the operation correction features; adjust the prediction weights in the long short-term memory network and the gradient booster model corresponding to the adjustment instructions based on the matched adjustment instructions, so as to obtain the adjusted long short-term memory network and the adjusted gradient booster model.

[0173] In this embodiment, the prediction module 714 can also be used to perform... Figure 3 For a detailed description of the prediction module 714, see steps A1-A3 shown below.

[0174] Specifically, the system further includes a processing scheme generation module 715, which is used to generate a fault report corresponding to the industrial equipment based on the first prediction result and / or the second prediction result and / or the third prediction result.

[0175] This application also provides an electronic device, the structure of which is as follows: Figure 5As shown, the electronic device includes a memory 811, a processor 812, a communication module 813, and an input / output interface 814, etc. Optionally, the memory 811, the processor 812, the communication module 813, and the input / output interface 814 can be connected and communicate with each other through a bus 815.

[0176] The memory 811 is used to store one or more computer programs and to transfer the code of the computer programs to the processor 812; when the one or more computer programs are executed by the processor 812, an industrial equipment fault prediction method in this application embodiment is implemented.

[0177] Optionally, the electronic device can be connected to a network via communication module 813 to communicate with other devices, such as terminals or servers, and to interact with data. The electronic device can be various forms of digital computers, exemplarily such as desktop computers, servers, workbenches, mainframes, or other types of computers. The electronic device can also be various forms of mobile terminals, exemplarily such as smartphones, tablets, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0178] Optionally, the electronic device can connect to desired input / output devices, such as a keyboard or display device, via the input / output interface 814. The electronic device itself may have a display device, and other display devices can also be connected externally via the input / output interface 814. Optionally, a storage device, such as a hard disk, can also be connected via the input / output interface 814 to store data from the electronic device, read data from the storage device, or store data from the storage device in the memory 811. It is understood that the input / output interface 814 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 814 can be a component of the electronic device or an external device connected to the electronic device when needed.

[0179] Optionally, the memory 811 may be a volatile memory and / or a non-volatile memory. The volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0180] Optionally, the computer program stored in the memory 811 can be divided into one or more modules, which are stored in the memory 811 and executed by the processor 812 to perform the method provided in this embodiment. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0181] Optionally, the processor 812 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 812 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and can also be any suitable controller, microcontroller, processor, etc. The processor 812 executes the various methods and processes of this embodiment, exemplified by, an industrial equipment fault prediction method according to an embodiment of this application.

[0182] Optionally, the bus 815 may include a path for transmitting information. Depending on its function, the bus 815 may be divided into an address bus, a data bus, a control bus, etc.

[0183] In an optional implementation, this application embodiment also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the methods described in the above-described method embodiments. Part or all of the computer program may be loaded and / or installed on the memory 811 of an electronic device. When the computer program is executed by the processor 812, one or more steps of an industrial equipment fault prediction method according to an embodiment of this application can be performed.

[0184] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0185] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for predicting industrial equipment failures, characterized in that, The method includes: Acquire operational and environmental data of the industrial equipment, and acquire historical fault data and spare parts data related to the industrial equipment. The operating data is subjected to feature extraction to obtain operating features; the historical fault data is subjected to feature extraction to obtain historical fault association features; and the spare parts data is subjected to feature extraction to obtain spare parts association features. The environmental data is input into a linear environmental disturbance compensation model to obtain a correction amount. The operating characteristics are then corrected based on the correction amount to obtain a first intermediate correction feature. The environmental data and the operating characteristics are input into a nonlinear environmental disturbance separation model to obtain a second intermediate correction feature. The first intermediate correction feature and the second intermediate correction feature are then weighted and fused according to preset weights to obtain the operating correction feature. Acquire maintenance knowledge data corresponding to industrial equipment and convert the maintenance knowledge data into several adjustment instructions; acquire adjustment instructions that match the operation correction features; adjust the prediction weights in the preset long short-term memory network and gradient booster model that correspond to the preset adjustment instructions based on the matched adjustment instructions, to obtain the adjusted long short-term memory network and the adjusted gradient booster model. The operational correction features are input into the adjusted long short-term memory network to obtain the first prediction result; The operational correction features, the first prediction result, the historical fault association features, and the spare parts association features are input into the adjusted gradient booster model to obtain the second prediction result. The first prediction result and the second prediction result are input into a preset industrial large model to obtain a third prediction result.

2. The method according to claim 1, characterized in that, Before the step of extracting features from the running data to obtain running features, the method further includes: Obtain the location points corresponding to the missing data in the running data; When the number of points at the locations of several consecutive missing data is less than or equal to a preset number threshold, the missing data is filled in using interpolation to obtain the filled running data, and outliers in the filled running data are removed based on preset rules to obtain the preprocessed running data. When the number of points with consecutive missing data exceeds the threshold, the running data is discarded, and the missing data situation is reported. The step of extracting features from the running data to obtain running features specifically involves extracting features from the preprocessed running data to obtain running features.

3. The method according to claim 1, characterized in that, The method further includes: Several fault severity levels and several fault types are preset; The first prediction result includes the predicted occurrence time of the industrial equipment failure and the severity level of the failure matched by the industrial equipment; the second prediction result includes the failure type and failure location matched by the industrial equipment; and the third prediction result includes the failure diagnosis, failure handling plan and spare parts allocation plan for the industrial equipment.

4. The method according to any one of claims 1-3, characterized in that, The method further includes: Based on a preset algorithm, feature filtering is performed on the operation correction features, the historical fault association features, and the spare parts association features to obtain the operation core features, the historical fault association core features, and the spare parts association core features. The step of inputting the operational correction features into the adjusted long short-term memory network to obtain the first prediction result specifically involves: inputting the operational core features into the adjusted long short-term memory network to obtain the first prediction result; The second prediction result is obtained by inputting the operational correction features, the first prediction result, the historical fault association features, and the spare parts association features into the adjusted gradient booster model. Specifically, the second prediction result is obtained by inputting the operational core features, the first prediction result, the historical fault association core features, and the spare parts association core features into the adjusted gradient booster model.

5. The method according to claim 1, characterized in that, After the step of inputting the first prediction result and the second prediction result into a preset industrial large-scale model to obtain the third prediction result, the method further includes: A fault report corresponding to the industrial equipment is generated based on the first prediction result and / or the second prediction result and / or the third prediction result.

6. An industrial equipment fault prediction system, characterized in that, The system includes: The data acquisition module is used to acquire the operating data and environmental data of the industrial equipment, as well as the historical fault data and spare parts data related to the industrial equipment. The feature extraction module is used to extract features from the operating data to obtain operating features; to extract features from the historical fault data to obtain historical fault association features; and to extract features from the spare parts data to obtain spare parts association features. The correction module is used to input the environmental data into a linear environmental interference compensation model to obtain a correction amount, and to correct the operating characteristics according to the correction amount to obtain a first intermediate correction feature; input the environmental data and the operating characteristics into a nonlinear environmental interference separation model to obtain a second intermediate correction feature; and to perform weighted fusion of the first intermediate correction feature and the second intermediate correction feature according to a preset weight to obtain an operating correction feature. The prediction module is used to acquire maintenance knowledge data corresponding to industrial equipment and convert the maintenance knowledge data into several adjustment instructions; acquire adjustment instructions that match the operation correction features; adjust the prediction weights corresponding to the preset adjustment instructions in the preset Long Short-Term Memory Network and Gradient Boosting Machine Model based on the matched adjustment instructions to obtain the adjusted Long Short-Term Memory Network and the adjusted Gradient Boosting Machine Model; input the operation correction features into the adjusted Long Short-Term Memory Network to obtain a first prediction result; input the operation correction features, the first prediction result, the historical fault association features, and the spare parts association features into the adjusted Gradient Boosting Machine Model to obtain a second prediction result; and input the first prediction result and the second prediction result into the preset large industrial model to obtain a third prediction result.

7. The system according to claim 6, characterized in that, The prediction module also includes: Based on a preset algorithm, feature filtering is performed on the operation correction features, the historical fault association features, and the spare parts association features to obtain the operation core features, the historical fault association core features, and the spare parts association core features. The step of inputting the operational correction features into the adjusted long short-term memory network to obtain the first prediction result specifically involves: inputting the operational core features into the adjusted long short-term memory network to obtain the first prediction result; The second prediction result is obtained by inputting the operational correction features, the first prediction result, the historical fault association features, and the spare parts association features into the adjusted gradient booster model. Specifically, the second prediction result is obtained by inputting the operational core features, the first prediction result, the historical fault association core features, and the spare parts association core features into the adjusted gradient booster model.

8. The system according to claim 6, characterized in that, The system is also used for: A fault report corresponding to the industrial equipment is generated based on the first prediction result and / or the second prediction result and / or the third prediction result.

9. An electronic device, characterized in that, include: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements an industrial equipment fault prediction method as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute and implement the industrial equipment fault prediction method as described in any one of claims 1-5.

Images