An industrial device data repair method, system, device, medium and product

Abstract

The application discloses an industrial equipment data repairing method, system, device, medium and product, relates to the field of industrial equipment data repairing, and comprises the following steps: preprocessing multi-source data, and marking abnormal data in the preprocessed multi-source data; the multi-source data comprises sensor data and equipment working condition parameters; inputting the preprocessed multi-source data into a data repairing model, repairing the abnormal data, and generating repairing data; the data repairing model is constructed based on a conditional generative adversarial network; performing physical rationality verification on the repairing data, determining a verification result, adjusting a dynamic repairing strategy according to the abnormal data and the verification result, selecting a data repairing mode, and repairing the abnormal data according to the data repairing mode.

Description

Technical Field

[0001] This application relates to the field of industrial equipment data repair, and in particular to a method, system, equipment, medium and product for industrial equipment data repair. Background Technology

[0002] The current industrial equipment data restoration technology system mainly consists of statistical methods, machine learning models, deep learning methods, and physical model-driven methods. Statistical methods, such as linear interpolation and Autoregressive Integrated Moving Average (ARIMA) time series prediction, are widely used in the industrial field due to their ease of operation. However, they can only handle simple data loss or noise restoration and perform poorly in complex scenarios such as multi-sensor fusion and non-stationary signals. Machine learning models, such as support vector machines and random forests, rely on high-quality labeled data to identify abnormal patterns, but are limited by the scarcity of abnormal samples in industrial scenarios, resulting in insufficient model generalization ability. Deep learning methods, such as autoencoders and generative adversarial networks, can handle complex data distributions, but suffer from high computational resource requirements, poor real-time performance, and a lack of physical plausibility in the generated data. Physical model-driven methods combine equipment physical characteristics for data verification, but lack flexibility and are difficult to adapt to changes in various operating conditions. From the perspective of current industry applications, traditional methods still dominate, but their applicability in complex industrial environments is gradually showing bottlenecks. While deep learning technology has made progress at the academic level, it is difficult to achieve large-scale implementation due to the special constraints of industrial scenarios.

[0003] Existing technologies have revealed several key problems in practical applications.

[0004] Regarding repair accuracy and physical rationality: statistical methods cannot capture complex anomalies such as multi-sensor coupling faults, and data generated by deep learning may violate physical laws such as vibration energy conservation and temperature-current coupling. Furthermore, existing methods generally lack deep integration of operating parameters and sensor data, resulting in a disconnect between repair results and the actual state of the equipment.

[0005] In terms of real-time performance: the high computational resource requirements of complex models make it difficult to meet the millisecond-level response requirements of industrial equipment, and fixed repair strategies cannot be dynamically adjusted according to the severity of the anomaly, affecting repair efficiency. Summary of the Invention

[0006] The purpose of this application is to provide a method, system, device, medium, and product for industrial equipment data repair, in order to solve the problem of low efficiency in industrial equipment data repair.

[0007] To achieve the above objectives, this application provides the following solution.

[0008] In a first aspect, this application provides a method for repairing data in industrial equipment, comprising the following steps.

[0009] The multi-source data is preprocessed, and abnormal data in the preprocessed multi-source data is marked; the multi-source data includes sensor data and equipment operating parameters; the sensor data includes vibration triaxial signals, current, and temperature; the equipment operating parameters include load, rotational speed, and wear.

[0010] The preprocessed multi-source data is input into the data repair model to repair the abnormal data and generate repaired data; the data repair model is constructed based on conditional generative adversarial networks.

[0011] The physical rationality of the repaired data is verified, and the verification result is determined.

[0012] Adjust the dynamic repair strategy based on the abnormal data and the verification results, and select the data repair method.

[0013] The abnormal data is repaired according to the described data repair method.

[0014] Secondly, an industrial equipment data repair system includes the following modules.

[0015] The preprocessing module is used to preprocess multi-source data and mark abnormal data in the preprocessed multi-source data; the multi-source data includes sensor data and equipment operating parameters; the sensor data includes vibration triaxial signals, current and temperature; the equipment operating parameters include load, speed and wear.

[0016] The data repair generation module is used to input the preprocessed multi-source data into the data repair model to repair the abnormal data and generate repaired data; the data repair model is constructed based on conditional generative adversarial networks.

[0017] The verification module is used to verify the physical rationality of the repaired data and determine the verification result.

[0018] The selection module is used to adjust the dynamic repair strategy based on the abnormal data and the verification results, and to select the data repair method.

[0019] The repair module is used to repair the abnormal data according to the data repair method.

[0020] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described industrial equipment data repair method.

[0021] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described industrial equipment data repair method.

[0022] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described industrial equipment data repair method.

[0023] According to the specific embodiments provided in this application, this application has the following technical effects: This application uses a data repair model, combined with multi-source data such as equipment operating parameters and sensor data as input conditions, to generate repair data that is highly correlated with the equipment status, and performs physical rationality verification on the repair data to exclude repair data that violates natural laws, thereby improving data credibility and further improving repair accuracy and physical rationality.

[0024] Furthermore, this application adjusts the dynamic repair strategy based on abnormal data and verification results, thereby automatically selecting different data repair methods to repair abnormal data, achieving rapid repair of simple anomalies, avoiding complex calculations, meeting real-time requirements, and improving repair efficiency. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0026] Figure 1 This is a schematic flowchart of an industrial equipment data repair method provided in an embodiment of this application.

[0027] Figure 2 A schematic diagram of the data repair model structure provided in this application.

[0028] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] like Figure 1 As shown in the figure, this application provides a method for repairing data of industrial equipment, including the following steps.

[0032] S1: Preprocess the multi-source data and mark the abnormal data in the preprocessed multi-source data; the multi-source data includes sensor data and equipment operating parameters; the sensor data includes vibration triaxial signals, current and temperature; the equipment operating parameters include load, speed and wear.

[0033] S2: Input the preprocessed multi-source data into the data repair model to repair the abnormal data and generate repaired data; the data repair model is constructed based on conditional generative adversarial networks.

[0034] S3: Perform a physical validity check on the repaired data and determine the check result.

[0035] S4: Adjust the dynamic repair strategy based on the abnormal data and the verification results, and select a data repair method.

[0036] S5: Repair the abnormal data according to the data repair method.

[0037] In an exemplary embodiment, this application acquires vibration triaxial signals, current, temperature data, and equipment operating parameters in real time, aligns the timestamps, performs integrity verification, uses a dual mechanism of 3σ principle and physical constraints to mark abnormal data points, completes multi-source data preprocessing and anomaly localization, and uses cubic spline interpolation to fill missing values. S1 can be replaced by the following steps.

[0038] S11: Align the timestamps and fill in missing values ​​for the multi-source data in sequence to determine the preprocessed multi-source data.

[0039] S12: Set an anomaly detection threshold, and based on the anomaly detection threshold, use the 3σ principle and physical constraint dual mechanism to mark the abnormal data points in the preprocessed multi-source data to generate abnormal data.

[0040] In practical applications, the data preprocessing process specifically includes the following steps.

[0041] Aligning timestamps: This involves resampling and interpolating the time series data, resampling the input time series data down to the millisecond level, and filling the newly generated timestamps with the most recent valid data points. This time series data is thus multi-source data existing in time series form.

[0042] Specifically, the system collects vibration triaxial signals (i.e., X / Y / Z axis acceleration), current, temperature data, and equipment operating parameters (load, speed, wear) in real time. The computer system aligns the timestamps and samples the multi-source data to a unified 1-millisecond resolution. Missing timestamps are filled using the nearest neighbor interpolation method.

[0043] Set anomaly detection thresholds: Set thresholds for vibration acceleration, temperature, and current. Specifically, set thresholds for vibration acceleration (e.g., ±10g), temperature (e.g., 0~150℃), and current (e.g., 0~50A). Data points exceeding these thresholds are marked as anomalies.

[0044] Missing value imputation: Missing values ​​are imputed using cubic spline interpolation. Specifically, for data points marked as missing or invalid, cubic spline interpolation is used to impute them, ensuring temporal continuity.

[0045] In practical applications, the 3σ principle is to calculate the mean μ and standard deviation σ of each sensor data point and mark data points that exceed the range of [μ-3σ, μ+3σ] as anomalies.

[0046] This application provides high-confidence input data for subsequent repair by using multi-source data alignment and a dual anomaly detection mechanism, thereby reducing noise interference.

[0047] In one exemplary embodiment, this application constructs a data repair model (DataRepairGAN) based on a conditional generative adversarial network. Device operating parameters and sensor data are concatenated as input. The generator network contains three fully connected layers, employing LeakyReLU activation and layer normalization. The discriminator part distinguishes between real and generated data through adversarial training. Figure 2 As shown, the data repair model specifically includes a generator and a discriminator.

[0048] The generator comprises a first input layer, a first hidden layer, and a first output layer connected in sequence, used to determine generated data from random noise based on the preprocessed multi-source data. The first hidden layer comprises two layers: a first fully connected layer, a first activation function layer, and a first normalization layer connected in sequence; the first fully connected layer maps the preprocessed multi-source data input from the first input layer from 8 dimensions to 128 dimensions. The second hidden layer comprises a second fully connected layer, a second activation function layer, and a second normalization layer connected in sequence; the second fully connected layer maps the data output from the first normalization layer from 128 dimensions to 256 dimensions. The first output layer includes a third fully connected layer, used to map the data output from the second normalization layer from 256 dimensions to 5 dimensions. The generated data includes 5 features.

[0049] In practical applications, the generator is designed to generate samples from random noise, and its output is influenced by conditional information. Its architecture includes a first input layer, a first hidden layer, and a first output layer.

[0050] The first input layer has an input dimension of 8, representing 8 input features, including random noise and conditional information (such as load, rotation speed, wear, etc.).

[0051] The first hidden layer consists of two layers. The first layer includes a first fully connected layer: a fully connected layer maps the input from 8 dimensions to 128 dimensions; the first activation function layer uses the LeakyReLU activation function with a negative slope of 0.2, which can effectively prevent the "neuron death" problem; then a first normalization layer is used for layer normalization to improve training stability and accelerate convergence. The second layer includes a second fully connected layer: further maps the data from 128 dimensions to 256 dimensions; the second activation function layer also uses the LeakyReLU activation function with a negative slope of 0.2; then a second normalization layer is used for layer normalization again. The first output layer includes a third fully connected layer: mapping the 256-dimensional data to 5 dimensions, i.e., the repaired sensor data, indicating that the generated data will have 5 features; using the Tanh activation function, the output is compressed to the range [-1, 1].

[0052] The discriminator includes a second input layer, a second hidden layer, and a second output layer connected in sequence, used to distinguish between real data and generated data. The second input layer receives both real and generated data, with an 8-dimensional input. The second hidden layer comprises two layers: a first layer includes a fourth fully connected layer and a third activation function layer connected in sequence; the fourth fully connected layer maps the real and generated data input from the second input layer from 8 dimensions to 256 dimensions; the second layer includes a fifth fully connected layer and a fourth activation function layer connected in sequence; the fifth fully connected layer maps the data output from the third activation function layer from 256 dimensions to 128 dimensions; and the second output layer includes a sixth fully connected layer, used to map the data output from the fourth activation function layer from 128 dimensions to 1 dimension.

[0053] In practical applications, the discriminator architecture is used to distinguish between real samples and generated samples. The discriminator architecture is as follows: The second input layer has an input dimension of 8, representing the sample features entering the discriminator. These sample features include sample features from real data and sample features from generated data.

[0054] The first layer in the second hidden layer includes a fourth fully connected layer: mapping the input from 8 dimensions to 256 dimensions; the third activation function layer uses the LeakyReLU activation function with a negative slope of 0.2 to increase the non-linearity of the network; the second layer includes a fifth fully connected layer: mapping the data from 256 dimensions to 128 dimensions; the fourth activation function layer uses the LeakyReLU activation function.

[0055] The second output layer includes a sixth fully connected layer: finally, the 128-dimensional data is mapped to 1 dimension, and a scalar value is output to indicate the authenticity of the input sample; and the Sigmoid activation function is used to make the output value in the range of [0,1], representing the probability of the true sample.

[0056] In practical applications, the generator is as follows.

[0057] First input layer: 8-dimensional vector (5-dimensional sensor data + 3-dimensional operating parameters: load, speed, wear).

[0058] First hidden layer: 128-dimensional, LeakyReLU activation (negative slope 0.2), layer normalization. Second hidden layer: 256-dimensional, LeakyReLU activation (negative slope 0.2), layer normalization.

[0059] First output layer: 5-dimensional (repaired sensor data), compressed to [-1, 1] using the Tanh activation function.

[0060] The discriminator is as follows.

[0061] Second input layer: 8 dimensions (real or generated data + operating parameters).

[0062] Second hidden layer: First layer: 256 dimensions, LeakyReLU activation (negative slope 0.2). Second layer: 128 dimensions, LeakyReLU activation (negative slope 0.2).

[0063] Second output layer: 1-dimensional, Sigmoid activation function outputs the probability of truth.

[0064] The generator learns complex data distributions through adversarial training, while the discriminator improves the authenticity of the generated data, ensuring that the repaired data conforms to the actual operating conditions of the equipment.

[0065] In practical applications, the training parameters are configured as follows.

[0066] The learning rate is used in the Adam optimizer to control the step size of model updates. A smaller learning rate can improve stability and convergence. It is set to 0.0001.

[0067] Each batch contains 32 samples, meaning 32 samples are processed in one iteration, improving training efficiency.

[0068] Set the gradient clipping threshold to 1.0 to prevent gradient explosion and maintain training stability.

[0069] The condition dimension is set to 3, corresponding to the three important features in the input: load, speed, and wear.

[0070] In practical applications, the adversarial training configuration is as follows.

[0071] Optimizer: Adam (learning rate 0.0001, first moment (momentum) decay rate β1 = 0.5, second moment (adaptive learning rate) decay rate β2 = 0.999).

[0072] Training parameters: batch size 32, gradient clipping threshold 1.0, training epochs 500.

[0073] Condition information: Operating parameters (load, speed, wear) serve as additional inputs to the generator and discriminator.

[0074] In an exemplary embodiment, the physical dimensions of the recovered data are processed by an inverse normalization method to perform physical rationality verification. After processing, a fifth-order sliding window weighted average is used to achieve time series smoothing. S3 can be replaced by the following steps.

[0075] S31: Based on different physical constraints, perform physical rationality verification on the repaired data and determine the verification result; the physical constraints include vibration energy conservation constraints, kinematic derivative constraints, and temperature-current coupling constraints; among them, vibration energy conservation constraints are used to ensure that the energy consumed by the system during vibration does not exceed the rated energy, kinematic derivative constraints are used to limit the vibration change rate to avoid excessive vibration changes leading to system instability, and temperature-current coupling constraints ensure that the system does not generate excessive heat during operation.

[0076] The vibration energy conservation constraint is: ;in, m Equivalent mass represents the mass components involved in an industrial equipment system. The speed signal of equipment vibration. k E is the stiffness coefficient, used to measure the elastic properties of a system. 额定 Rated energy, which is the upper limit of energy set during system design. x The displacement signal represents the vibration of the equipment; the temperature-current coupling constraint is: ; where Δ T Δ is the change in temperature. t For the time change, IThis is the motor current. R This is the equivalent thermal resistance.

[0077] In practical applications, physical rationality verification is preceded by data denormalization and time-series smoothing.

[0078] Specifically, the normalized data ([-1,1]) output by the generator is denormalized to the original physical dimensions (such as acceleration in g).

[0079] The time series data were smoothed using a five-order sliding window weighted average method (window length 5ms, weight coefficients [0.1, 0.2, 0.4, 0.2, 0.1]).

[0080] This application uses physical constraint verification to exclude repair results that violate natural laws, thereby improving data credibility.

[0081] In one exemplary embodiment, S4 can be replaced by the following steps.

[0082] S41: Determine the statistical anomaly score based on the 3σ principle according to the abnormal data.

[0083] S42: Determine the violation score based on the physical constraints according to the verification results.

[0084] S43: Determine the anomaly score based on the statistical anomaly score and the violation score based on physical constraints.

[0085] S44: Adjust the dynamic repair strategy based on the anomaly score to determine the data repair method.

[0086] In practical applications, the anomaly score S = alpha cdot S_{text{statistics}} + beta cdot S_{text{physics}}).

[0087] S_{text{statistics}}: Statistical anomaly score (0~1) based on the 3σ principle.

[0088] S_{text{physics}}: Score (0~1) for the degree of violation based on physical constraints.

[0089] alpha, beta: weighting coefficients, alpha=0.6, beta=0.4.

[0090] cdot: Multiplication operation.

[0091] In an exemplary embodiment, in practical applications, the dynamic repair strategy is adjusted, and the repair level is automatically selected based on the anomaly score, namely the linear interpolation method, the ARIMA prediction method, and the data repair model. The model is updated online based on the incremental learning mechanism. S44 can be replaced by the following steps.

[0092] S441: When the abnormal score belongs to the first abnormal score interval, the linear interpolation method is selected as the data repair method.

[0093] S442: When the abnormal score belongs to the second abnormal score interval, the ARIMA prediction method is selected as the data repair method.

[0094] S443: When the abnormal score belongs to the third abnormal score interval, select the data repair model as the data repair method.

[0095] In practical applications, the dynamic repair strategy is determined based on the anomaly score range. The specific rules of the dynamic repair strategy are shown in Table 1.

[0096] Table 1

[0097]

[0098] Among them, linear interpolation is used for rapid repair of transient noise with a short response time. ARIMA prediction is applied to short-term data loss by predicting future values ​​to fill in the gaps. DataRepairGAN, the data repair model constructed in this application, is suitable for the generation of anomalies in complex patterns, with a relatively long response time, but it can generate high-quality repaired data.

[0099] As shown in Table 1, when S∈[0,0.3), it is considered to be instantaneous noise, so the lowest cost linear interpolation method is used to repair the data, and the response time is <1ms.

[0100] When S∈[0.3, 0.7), the ARIMA prediction method is used to handle short-term data loss, and the response time is <5ms.

[0101] When S∈[0.7, 1.0], DataRepairGAN is called to repair complex patterns, with a response time of <10ms.

[0102] It is evident that this dynamic repair strategy employs a hierarchical mechanism, which balances the requirements for repair quality and real-time performance.

[0103] In an exemplary embodiment, S5 can be replaced by the following steps.

[0104] S51: Monitor the weight change rate of the data repair model and update the data repair model according to the incremental learning mechanism; the incremental update formula for the model parameters is: ;in, for t Model parameters at time +1, for t Model parameters at time 10:00 η For incremental learning rate, x new This serves as the new training sample, i.e., the preprocessed multi-source data.

[0105] In practical applications, this application updates the model online based on an incremental learning mechanism, which includes the following steps to ensure the model's continuous adaptability and accuracy.

[0106] Concept drift is detected by monitoring changes in model parameters (i.e., the rate of change).

[0107] ;in, W t and W t-1 These are the model weights for the current and previous training iterations. Retraining is triggered when the weight change rate exceeds 0.1.

[0108] Regarding adaptability, sensor characteristic drift or environmental changes during long-term operation of industrial equipment can lead to data distribution shifts. Traditional models, lacking incremental update mechanisms, require frequent full retraining, resulting in high maintenance costs. This application, however, dynamically updates model parameters by monitoring the model weight change rate and employing an incremental learning mechanism. This avoids performance degradation caused by changes in equipment status, thus adapting to long-term conceptual drift issues such as equipment aging and environmental changes, and maintaining the stability of the repair effect.

[0109] In the anomaly detection stage, detection methods relying solely on statistical thresholds (such as 3σ) have a high false alarm rate in non-Gaussian distributed data, are susceptible to transient noise interference such as peak values ​​during equipment start-up and shutdown, and lack robustness. This application provides a joint labeling method combining the 3σ principle and physical thresholds. By combining statistical anomalies (i.e., the 3σ range) with physical thresholds (vibration / temperature / current exceeding limits), it avoids the limitations of a single detection mechanism, thereby significantly reducing the false alarm and false negative rates through a dual anomaly detection mechanism (statistical + physical).

[0110] Furthermore, traditional methods lack physical constraints, and the repaired data may violate the operating principles of the equipment (such as a mismatch between temperature rise and current). In contrast, this application enforces the verification of data rationality through physical constraints. The repaired data strictly meets the physical characteristics of industrial equipment (such as energy conservation and thermodynamic relationships), and can be directly used for equipment health assessment and control decisions.

[0111] This application also provides an industrial equipment data repair system, including the following modules.

[0112] The preprocessing module is used to preprocess multi-source data and mark abnormal data in the preprocessed multi-source data; the multi-source data includes sensor data and equipment operating parameters; the sensor data includes vibration triaxial signals, current and temperature; the equipment operating parameters include load, speed and wear.

[0113] The data repair generation module is used to input the preprocessed multi-source data into the data repair model to repair the abnormal data and generate repaired data; the data repair model is constructed based on conditional generative adversarial networks.

[0114] The verification module is used to verify the physical rationality of the repaired data and determine the verification result.

[0115] The selection module is used to adjust the dynamic repair strategy based on the abnormal data and the verification results, and to select the data repair method.

[0116] The repair module is used to repair the abnormal data according to the data repair method.

[0117] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device may be a server or a terminal, and its internal structure diagram may be as follows: Figure 3 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data to be processed. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a data repair method for industrial equipment.

[0118] Those skilled in the art will understand that Figure 3The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0119] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0120] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0121] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0123] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0124] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for data repair of industrial equipment, characterized in that, include: Preprocess the multi-source data and mark outliers in the preprocessed multi-source data; The multi-source data includes sensor data and equipment operating parameters; the sensor data includes triaxial vibration signals, current, and temperature; the equipment operating parameters include load, rotational speed, and wear. The preprocessed multi-source data is input into the data repair model to repair the abnormal data and generate repaired data. The data repair model is built on a conditional generative adversarial network; the data repair model specifically includes a generator and a discriminator. The generator comprises a first input layer, a first hidden layer, and a first output layer connected in sequence, used to determine generated data from random noise based on the preprocessed multi-source data; wherein, the first hidden layer comprises two layers, the first layer comprising a first fully connected layer, a first activation function layer, and a first normalization layer connected in sequence; the first fully connected layer is used to map the preprocessed multi-source data input from the first input layer from 8 dimensions to 128 dimensions; the second layer comprises a second fully connected layer, a second activation function layer, and a second normalization layer connected in sequence; the second fully connected layer is used to map the data output from the first normalization layer from 128 dimensions to 256 dimensions; the first output layer comprises a third fully connected layer, used to map the data output from the second normalization layer from 256 dimensions to 5 dimensions; the generated data comprises 5 features; The discriminator comprises a second input layer, a second hidden layer, and a second output layer connected in sequence, used to distinguish between real data and generated data. The second input layer receives both real and generated data, with an 8-dimensional input. The second hidden layer consists of two layers: a first layer comprises a fourth fully connected layer and a third activation function layer connected in sequence; the fourth fully connected layer maps the real and generated data input from the second input layer from 8 dimensions to 256 dimensions; the second layer comprises a fifth fully connected layer and a fourth activation function layer connected in sequence; the fifth fully connected layer maps the data output from the third activation function layer from 256 dimensions to 128 dimensions; and the second output layer comprises a sixth fully connected layer, used to map the data output from the fourth activation function layer from 128 dimensions to 1 dimension. The physical rationality of the repaired data is verified, and the verification result is determined. Adjust the dynamic repair strategy based on the abnormal data and the verification results, and select a data repair method; The abnormal data is repaired according to the described data repair method.

2. The industrial equipment data repair method according to claim 1, characterized in that, Preprocessing of multi-source data and marking of outliers in the preprocessed multi-source data, specifically including: The multi-source data is sequentially aligned with timestamps and missing values ​​are filled to determine the preprocessed multi-source data; An anomaly detection threshold is set, and based on the anomaly detection threshold, anomaly data points in the preprocessed multi-source data are marked using a dual mechanism of the 3σ principle and physical constraints to generate anomaly data; where σ is the standard deviation.

3. The industrial equipment data repair method according to claim 1, characterized in that, The physical rationality of the repaired data is verified, and the verification result is determined, specifically including: Based on different physical constraints, the physical rationality of the repaired data is verified, and the verification result is determined. The physical constraints include vibration energy conservation constraints, kinematic derivative constraints, and temperature-current coupling constraints. The vibration energy conservation constraint is as follows: ;in, m For equivalent quality, The speed signal of equipment vibration. k E is the stiffness coefficient. 额定 Rated energy, x The displacement signal represents the vibration of the equipment; the temperature-current coupling constraint is: ; where Δ T Δ is the change in temperature. t For the time change, I This is the motor current. R This is the equivalent thermal resistance.

4. The industrial equipment data repair method according to claim 1, characterized in that, The dynamic repair strategy is adjusted based on the abnormal data and the verification results, and a data repair method is selected, specifically including: A statistical anomaly score based on the 3σ principle is determined based on the anomaly data; A violation score based on physical constraints is determined based on the verification results; An anomaly score is determined based on the statistical anomaly score and the violation score based on physical constraints; The dynamic repair strategy is adjusted based on the anomaly score to determine the data repair method, specifically including: When the abnormal score belongs to the first abnormal score interval, linear interpolation method is selected as the data repair method; When the abnormal score belongs to the second abnormal score interval, the ARIMA prediction method is selected as the data repair method. When the abnormal score belongs to the third abnormal score interval, the data repair model is selected as the data repair method.

5. The industrial equipment data repair method according to claim 1, characterized in that, The abnormal data is repaired according to the described data repair method, and then the process further includes: Monitor the weight change rate of the data repair model and update the data repair model according to the incremental learning mechanism; the incremental update formula for the model parameters is: ;in, for t Model parameters at time +1, for t Model parameters at time 10:00 η For incremental learning rate, x new This serves as the new training sample, i.e., the preprocessed multi-source data.

6. An industrial equipment data repair system, characterized in that, The industrial equipment data repair system employs the industrial equipment data repair method according to any one of claims 1-5, and the industrial equipment data repair system comprises: The preprocessing module is used to preprocess multi-source data and mark abnormal data in the preprocessed multi-source data; the multi-source data includes sensor data and equipment operating parameters; the sensor data includes vibration triaxial signals, current and temperature; the equipment operating parameters include load, rotational speed and wear. The data repair generation module is used to input the preprocessed multi-source data into the data repair model to repair the abnormal data and generate repaired data; the data repair model is constructed based on a conditional generative adversarial network. The verification module is used to verify the physical rationality of the repaired data and determine the verification result. The selection module is used to adjust the dynamic repair strategy based on the abnormal data and the verification results, and to select the data repair method. The repair module is used to repair the abnormal data according to the data repair method.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the industrial equipment data repair method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the industrial equipment data repair method according to any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the industrial equipment data repair method according to any one of claims 1-5.

Images