Comprehensive evaluation method and system for PHM system of electromechanical system

Abstract

The invention relates to the technical field of electromechanical system evaluation, and discloses a comprehensive evaluation method and system for an electromechanical system PHM system, and the method comprises the steps: collecting the multi-source operation data of an electromechanical system, and carrying out the preprocessing and feature extraction; a fault diagnosis model based on a deep neural network DNN, a health state evaluation model based on a random forest RF and a residual life prediction model based on a particle filter PF are constructed, and fault classification, health state recognition and residual life estimation are performed on the electromechanical system. Then, by defining a multi-model fusion algorithm, performing comprehensive analysis on output results of all the models, and obtaining a comprehensive health assessment result of the system; and finally, generating an evaluation report containing the fault type, the health state, the residual life and the maintenance suggestion. According to the invention, efficient and intelligent maintenance of the electromechanical system is realized, and the accuracy of fault prediction, the fineness of health state evaluation and the precision of residual life prediction are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electromechanical system evaluation, and particularly to a comprehensive evaluation method and system for a Prognostics and Health Management (PHM) system of an electromechanical system. Background Art

[0002] In the modern industrial field, as the core component of various equipment and production lines, the reliability and stability of electromechanical systems are directly related to production efficiency, product quality, and operation and maintenance costs. With the rapid development of intelligent manufacturing, higher requirements are put forward for the health management of electromechanical systems, that is, not only to be able to achieve rapid response and repair of faults, but also to have the ability of predictive maintenance to reduce unplanned downtime, extend equipment life, and optimize maintenance costs. Therefore, the Prognostics and Health Management (PHM) system has emerged as the key technology to improve the comprehensive performance of electromechanical systems.

[0003] The existing maintenance strategies for electromechanical systems are mostly based on run-to-failure or time-based maintenance. These methods often have problems of over-maintenance or under-maintenance, and it is difficult to effectively balance maintenance costs and equipment reliability. With the rapid development of technologies such as sensor technology, big data analysis, and artificial intelligence, powerful technical support has been provided for the PHM of electromechanical systems. However, the current PHM system still faces many challenges in practical applications:

[0004] Insufficient data fusion and processing capabilities: The data generated during the operation of electromechanical systems is huge and diverse in type, including sensor data, equipment logs, maintenance records, and operating environment data, etc. How to efficiently integrate these data and extract valuable features for fault prediction and health assessment is a major difficulty in current technologies.

[0005] Limited accuracy of fault diagnosis: Traditional fault diagnosis methods often rely on expert experience or simple rule matching, and it is difficult to handle complex and changeable fault patterns. Although the fault diagnosis models based on machine learning have made progress, they still have limitations in dealing with non-linear and high-dimensional data.

[0006] Lack of refinement in health state assessment: The health state of an electromechanical system is a dynamically changing process. Accurately assessing its current state and future trend is crucial for formulating maintenance strategies. Most of the existing assessment methods are based on single indicators or simple models, and it is difficult to comprehensively reflect the true health status of the system.

[0007] Low accuracy of remaining useful life prediction: Remaining useful life prediction is one of the core functions of the PHM system. However, due to factors such as data quality, model complexity, and algorithm accuracy, existing prediction methods often fail to achieve ideal results, especially when dealing with long-cycle and small-sample data.

[0008] Insufficient multi-model fusion and decision support: The PHM system involves multiple sub-models (such as fault diagnosis, health assessment, life prediction, etc.). How to effectively fuse the output results of these models to form a unified and accurate comprehensive health assessment and provide a scientific basis for maintenance decisions is a hot and difficult issue in current research. Summary of the Invention

[0009] The purpose of the present invention is to provide a comprehensive evaluation method and system for the PHM system of electromechanical systems to solve the problems raised in the above background technology.

[0010] To achieve the above purpose, the present invention provides the following technical solution: A comprehensive evaluation method for the PHM system of electromechanical systems, the method includes;

[0011] F1: Collect the operation data of the electromechanical system, including sensor data, equipment logs, maintenance records, and operating environment data; preprocess the collected data, including data cleaning, format conversion, time synchronization, and outlier detection, and then perform feature extraction to form a feature data set;

[0012] F2: Build a fault diagnosis model based on the deep neural network DNN algorithm. This model is trained using the sensor data and equipment logs in the feature data set to classify the fault types of the electromechanical system;

[0013] F3: Build a health status assessment model based on the random forest RF algorithm. This model uses the preprocessed feature data set to learn and identify the patterns of different health states;

[0014] F4: Build a remaining useful life prediction model based on the particle filter PF algorithm. This model uses historical maintenance records and operating environment data to predict the remaining useful life of the key components of the electromechanical system;

[0015] F5: Input the feature data set into the fault diagnosis model, health status assessment model, and remaining useful life prediction model respectively to obtain the fault diagnosis result, health status assessment result, and remaining useful life prediction result of the electromechanical system;

[0016] F6: Define a multi-model fusion algorithm to perform fusion analysis on the obtained model output results to obtain a comprehensive health assessment result;

[0017] F7: Generate an evaluation report for the electromechanical system PHM system based on the comprehensive health assessment results, including the fault type, health status, remaining life, and corresponding maintenance suggestions.

[0018] Preferably, the sensor data includes temperature, pressure, vibration, and current data; the equipment logs include the running time, number of shutdowns, and fault history; the maintenance records include the maintenance time, maintenance type, and maintenance cost; the operating environment data includes the ambient temperature, humidity, and dust concentration.

[0019] Preferably, the fault diagnosis model adopts a multi-layer perceptron (MLP) structure, extracts features through multiple fully connected layers, and finally outputs the fault type through a Softmax layer. Its network structure includes:

[0020] Input layer: Receive the preprocessed sensor data and equipment logs. The data dimension is unified as [N, F], where N is the number of samples and F is the number of features.

[0021] Hidden layer: Adopt multiple fully connected operations. After each full connection, a ReLU activation function is connected to extract features.

[0022] Output layer: Use the Softmax function to output the probability of each fault type. The number of output nodes is the number of fault types.

[0023] Preferably, the training steps of the fault diagnosis model include:

[0024] F201: Initialize the parameters of the DNN model, including weights, bias terms, and learning rate.

[0025] F202: Divide the preprocessed feature dataset into a training set and a validation set. The training set is used for model training, and the validation set is used for model validation.

[0026] F203: Use the training set to perform iterative training on the DNN model. In each iteration, in the forward propagation, calculate the model output, and calculate the loss between the predicted value and the true value according to the loss function; the loss function adopts the cross-entropy loss function.

[0027] F204: Update the model parameters according to the loss value through the backpropagation algorithm and the gradient descent method; the formula for updating the model parameters by the gradient descent method is:

[0028]

[0029] Where represents the model parameters, including all weights and biases that the model needs to learn; represents the learning rate, which is used to control the step size of parameter update; represents the loss function Regarding parameters , the gradient is a vector pointing in the direction of the steepest increase of the loss function; represents an assignment operation for updating the parameter value;

[0030] F205: Repeat steps F103 to F104 until the loss converges or the preset number of iterations is reached;

[0031] F206: Use the validation set to validate the trained DNN model and evaluate the classification accuracy of the model;

[0032] F207: When the model performance reaches the preset standard, save the model parameters to obtain the trained fault diagnosis model.

[0033] Preferably, the health status assessment model constructs a random forest using multiple decision trees and outputs the health status assessment result through a voting mechanism. Its structure includes:

[0034] Input layer: Receive the preprocessed feature data set;

[0035] Decision tree layer: Construct multiple decision trees, and each tree independently classifies the input data;

[0036] Output layer: Adopt a voting mechanism to comprehensively combine the classification results of all decision trees and output the health status assessment result.

[0037] Preferably, the steps for training the health status assessment model include:

[0038] F301: Use the preprocessed feature data set as the training set and select the health status as the label;

[0039] F302: Determine the number and maximum depth of the decision trees in the random forest;

[0040] F303: Use the training set to train the random forest model until the model converges;

[0041] F304: Use the validation set to validate the random forest model and evaluate the classification accuracy and generalization ability of the model;

[0042] F305: When the model performance reaches the preset standard, save the model parameters to obtain the trained health status assessment model.

[0043] Preferably, the remaining useful life prediction model adopts a particle filter algorithm to recursively estimate the remaining useful life of key components. Its implementation steps include:

[0044] F401: Screen out the historical maintenance records and operating environment data of key components from the preprocessed feature dataset to form a time series dataset;

[0045] F402: Define a state space model, including a state transition equation and an observation equation;

[0046] F403: Initialize the particle set, where each particle represents a possible value of the remaining life of the key component;

[0047] F404: Recursively update the particle set according to the state transition equation and the observation equation to obtain an estimated value of the remaining life;

[0048] F405: Use the validation set to validate the model and evaluate the prediction accuracy and stability of the model;

[0049] F406: When the model performance reaches the preset standard, save the model parameters to obtain a trained remaining life prediction model.

[0050] Preferably, the specific implementation steps of defining the multi-model fusion algorithm in step F6 include:

[0051] F601: Define the fault type output by the fault diagnosis model as D_DNN, the health status evaluation result output by the health status evaluation model as D_RF, and the remaining life output by the remaining life prediction model as D_PF;

[0052] F602: For D_DNN and D_RF, if the outputs of the two are inconsistent, perform weighted averaging according to the confidence or accuracy of their respective models to obtain the fused health status evaluation result D_Fusion_Health;

[0053] F603: Combine D_PF with the fused health status evaluation result D_Fusion_Health to form a comprehensive health evaluation result including the current health status and future remaining life;

[0054] F604: If there are multiple key components, perform the above fusion process for each component respectively to obtain their respective comprehensive evaluation results;

[0055] F605: According to the comprehensive evaluation results of all key components, use the weighted average method to obtain the final comprehensive health evaluation result D_Final;

[0056] F606: Compare D_Final with the preset health standard or threshold to determine the overall health status, potential faults and remaining life of the electromechanical system.

[0057] Preferably, in the evaluation report generated in step S7, the comprehensive evaluation results are visually displayed, including drawing a health status trend chart, a remaining life distribution chart, and a failure type proportion chart.

[0058] Preferably, a comprehensive evaluation system for an electromechanical system PHM system, the system includes:

[0059] A data collection and preprocessing module, used to collect the operation data of the electromechanical system, including sensor data, equipment logs, maintenance records, and operating environment data, and preprocess the collected data, including data cleaning, format conversion, time synchronization, and outlier detection, and then perform feature extraction to form a feature data set;

[0060] A fault diagnosis model construction module, used to construct a fault diagnosis model based on the deep neural network DNN algorithm, and this model is trained using the sensor data and equipment logs in the feature data set to classify the fault types of the electromechanical system;

[0061] A health status evaluation model construction module, used to construct a health status evaluation model based on the random forest RF algorithm, and this model uses the preprocessed feature data set to identify the patterns of different health statuses;

[0062] A remaining life prediction model construction module, used to construct a remaining life prediction model based on the particle filter PF algorithm, and this model uses historical maintenance records and operating environment data to predict the remaining life of the key components of the electromechanical system;

[0063] A model output acquisition module, used to input the feature data set into the fault diagnosis model, the health status evaluation model, and the remaining life prediction model respectively, to obtain the fault diagnosis result, the health status evaluation result, and the remaining life prediction result of the electromechanical system;

[0064] A multi-model fusion analysis module, used to define a multi-model fusion algorithm, and perform fusion analysis on the output results obtained by the model output acquisition module to obtain a comprehensive health evaluation result;

[0065] An evaluation report generation module, which is used to generate an evaluation report for the PHM system of the electromechanical system according to the comprehensive health evaluation results, including fault types, health status, remaining life, and corresponding maintenance suggestions. Compared with the prior art, the beneficial effects of the present invention are as follows: By collecting multi-dimensional operation data of the electromechanical system, including sensor data, equipment logs, maintenance records, and operation environment data, and adopting advanced data preprocessing technologies, such as data cleaning, format conversion, time synchronization, and outlier detection, the present invention effectively solves the problem of insufficient data fusion and processing capabilities. It improves the accuracy and reliability of the data, providing a solid foundation for subsequent fault prediction and health evaluation. By using the deep neural network (DNN) algorithm to construct a fault diagnosis model, the present invention can automatically learn and identify complex fault patterns, significantly improving the accuracy of fault diagnosis. Compared with traditional methods based on expert experience or simple rule matching, the DNN model can process non-linear and high-dimensional data, and is more adaptable to the complexity and diversity of modern electromechanical systems. By constructing a health status evaluation model based on the random forest (RF) algorithm, the present invention can comprehensively consider multiple health indicators, learn and identify patterns of different health states. This comprehensive evaluation method not only reflects the current state of the electromechanical system, but also can predict its future trends, providing strong support for formulating scientific maintenance strategies.

[0066] The present invention uses the particle filter (PF) algorithm to construct a remaining life prediction model. Combining historical maintenance records and operation environment data, it can accurately predict the remaining life of key components of the electromechanical system. The present invention defines a multi-model fusion algorithm, which performs fusion analysis on the output results of the fault diagnosis model, health status evaluation model, and remaining life prediction model to obtain a comprehensive health evaluation result. This fusion method not only improves the accuracy of the evaluation, but also provides a scientific basis for maintenance decision-making, realizing the intelligent and refined maintenance of the electromechanical system. According to the comprehensive health evaluation results, the present invention can automatically generate an evaluation report for the PHM system of the electromechanical system, including fault types, health status, remaining life, and corresponding maintenance suggestions. This provides intuitive and comprehensive information support for maintenance personnel, helps to quickly respond to faults, formulate reasonable maintenance plans, and improve the overall operation and maintenance efficiency. Brief Description of the Drawings

[0067] Figure 1 It is the working principle diagram of a comprehensive evaluation method for a PHM system of an electromechanical system according to the present invention;

[0068] Figure 2 It is the training flow chart of the fault diagnosis model;

[0069] Figure 3 It is the training step diagram of the remaining life prediction model based on the particle filter algorithm. Detailed Embodiments

[0070] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0071] Please refer to Figures 1-3 , the present invention provides a technical solution: a comprehensive evaluation method for the PHM system of an electromechanical system, and the method includes:

[0072] F1. Data collection and preprocessing:

[0073] Collect various types of data generated during the operation of the electromechanical system, specifically including sensor data, equipment logs, maintenance records, and operating environment data. These data are collected in real time through industrial Internet of Things technology and stored in a database.

[0074] Preprocess the collected data. First, perform data cleaning to remove invalid, duplicate, and incorrect data; then perform format conversion to unify data from different sources into a processable format; then perform time synchronization to ensure that the timestamps of all data are consistent; finally, perform outlier detection to identify and process outliers in the data through statistical methods and machine learning algorithms. After the preprocessing is completed, feature extraction is performed to form a feature dataset, providing input for subsequent model training.

[0075] F2. Build a fault diagnosis model:

[0076] Select the deep neural network (DNN) algorithm as the basis for the fault diagnosis model. The DNN model has strong non-linear mapping ability and automatic feature learning ability, and is suitable for the recognition of complex fault patterns.

[0077] Use the sensor data and equipment logs in the preprocessed feature dataset as training samples to train the DNN model. During the training process, adjust the network structure, optimization algorithm, and parameter settings to improve the fault diagnosis accuracy of the model.

[0078] After the training is completed, input the feature dataset into the DNN model to obtain the fault diagnosis results of the electromechanical system, including information such as fault type, fault location, and fault degree.

[0079] F3. Build a health status evaluation model:

[0080] Select the random forest (RF) algorithm as the basis for the health status evaluation model. The RF algorithm has the advantages of processing high-dimensional data, strong anti-noise ability, and good generalization performance, and is suitable for the evaluation of the health status of electromechanical systems.

[0081] Use the preprocessed feature dataset as the training sample to train the RF model. During the training process, by adjusting parameters such as the number of decision trees and the maximum depth, improve the evaluation accuracy of the model.

[0082] After the training is completed, input the feature dataset into the RF model to obtain the health status evaluation results of the electromechanical system, including information such as the health status level and the health status trend.

[0083] F4. Construct the remaining useful life prediction model:

[0084] Select the particle filter (PF) algorithm as the basis of the remaining useful life prediction model. The PF algorithm is suitable for state estimation of nonlinear and non-Gaussian systems and can handle the uncertainty problems in the prediction of the remaining useful life of key components of the electromechanical system.

[0085] Utilize the historical maintenance records and operating environment data to construct a degradation model of the key components. Combine the degradation model with the PF algorithm to form the remaining useful life prediction model.

[0086] Input the feature dataset into the remaining useful life prediction model to obtain the remaining useful life prediction results of the key components of the electromechanical system, including information such as the remaining useful life value and the remaining useful life distribution.

[0087] F5. Multi-model fusion and comprehensive health assessment:

[0088] Define a multi-model fusion algorithm to conduct a fusion analysis on the fault diagnosis results, health status evaluation results, and remaining useful life prediction results. The fusion algorithm can adopt methods such as weighted average, Bayesian fusion, and fuzzy comprehensive evaluation, and select the most suitable fusion strategy according to the specific situation. Through the multi-model fusion algorithm, obtain the comprehensive health assessment results of the electromechanical system. The comprehensive health assessment results include information such as the fault type, health status, remaining useful life, and corresponding maintenance suggestions, which can comprehensively reflect the health status and maintenance requirements of the electromechanical system.

[0089] F6. Generate an evaluation report:

[0090] Generate an evaluation report for the PHM system of the electromechanical system according to the comprehensive health assessment results. The evaluation report displays information such as the fault type, health status, remaining useful life, and maintenance suggestions in the form of charts, texts, etc., which is convenient for maintenance personnel to understand and make decisions.

[0091] Store the evaluation report in the database and display it to the maintenance personnel through the user interface. The maintenance personnel can formulate a maintenance plan based on the evaluation report to carry out fault prevention and handling, thereby improving the reliability and stability of the electromechanical system.

[0092] The present invention will be further described below in conjunction with Embodiments 1 to 4:

[0093] Example 1:

[0094] This example describes the comprehensive evaluation of the electromechanical system PHM system, and adopts the deep neural network (DNN) algorithm with a multi-layer perceptron (MLP) structure for the fault diagnosis model. The detailed steps are as follows:

[0095] Collect various types of data generated during the operation of the electromechanical system, including:

[0096] Sensor data: temperature, pressure, vibration and current data;

[0097] Equipment logs: running time, number of shutdowns and fault history;

[0098] Maintenance records: maintenance time, maintenance type and maintenance cost;

[0099] Operating environment data: ambient temperature, humidity and dust concentration.

[0100] Preprocess the collected data. First, perform data cleaning to remove invalid, duplicate and incorrect data; then perform format conversion to ensure that all data formats are unified; then perform time synchronization to ensure that the timestamps of all data are consistent; finally, perform outlier detection and handle outliers in the data. After the preprocessing is completed, perform feature extraction to form a feature dataset, which provides input for subsequent model training.

[0101] The method for constructing the fault diagnosis model includes:

[0102] Select the deep neural network (DNN) algorithm with a multi-layer perceptron (MLP) structure as the basis of the fault diagnosis model. The network structure of the DNN model includes:

[0103] Input layer: Receive the preprocessed sensor data and equipment logs, and the data dimension is unified as [N, F], where N is the number of samples and F is the number of features;

[0104] Hidden layer: Adopt multi-layer fully connected operations, and each layer is followed by a ReLU activation function after full connection, which is used to extract features;

[0105] Output layer: Use the Softmax function to output the probability of each fault type, and the number of output nodes is the number of fault types.

[0106] The training steps of the fault diagnosis model include:

[0107] F201: Initialize the parameters of the DNN model, including weights, bias terms and learning rates;

[0108] F202: Divide the preprocessed feature dataset into a training set and a validation set. The training set is used for model training, and the validation set is used for model validation;

[0109] F203: Use the training set to iteratively train the DNN model. In each iteration, perform forward propagation to calculate the model output, and calculate the loss between the predicted value and the true value according to the cross-entropy loss function;

[0110] F204: Update the model parameters according to the loss value through the backpropagation algorithm and the gradient descent method. The formula for updating the model parameters by the gradient descent method is:

[0111] where, represents the model parameters, including all weights and biases that the model needs to learn; represents the learning rate, which is used to control the step size of parameter update; represents the loss function with respect to the parameter gradient of, is a vector pointing in the direction of the fastest increase of the loss function; represents the assignment operation for updating the value of the parameter ; F205: Repeat steps F203 to F204 until the loss converges or reaches the preset number of iterations; F206: Use the validation set to validate the trained DNN model and evaluate the classification accuracy of the model; F207: When the model performance reaches the preset standard, save the model parameters to obtain the trained fault diagnosis model.

[0112] Embodiment 2:

[0113] To achieve the health state assessment of the electromechanical system, the present invention uses the random forest algorithm to construct a health state assessment model. The specific implementation steps are as follows:

[0114] Data preparation and preprocessing: Collect relevant data during the operation of the electromechanical system and perform preprocessing. The preprocessing process includes data cleaning, format conversion, time synchronization, outlier detection and processing, and feature extraction. After preprocessing, a feature dataset is formed to provide input for subsequent model training.

[0115] Construct a health state assessment model. The health state assessment model uses the random forest algorithm, and its structure includes:

[0116] Input layer: Receive the preprocessed feature dataset;

[0117] Decision tree layer: Construct multiple decision trees, and each tree independently classifies the input data;

[0118] Output layer: Adopt a voting mechanism to comprehensively combine the classification results of all decision trees and output the health status assessment result.

[0119] The training steps of the health status assessment model include:

[0120] F301: Use the preprocessed feature dataset as the training set and select the health status as the label. The health status can be defined according to actual needs, such as normal, minor fault, severe fault, etc.

[0121] F302: Determine the number of decision trees and the maximum depth in the random forest. These two parameters have an important impact on the performance and effect of the model and need to be determined through experiments or experience.

[0122] F303: Use the training set to train the random forest model. During the training process, each decision tree will independently classify the input data and build a tree structure by continuously splitting nodes. The training process will continue until the model converges or reaches the preset number of iterations.

[0123] F304: Use the validation set to validate the random forest model. The validation set is a dataset independent of the training set and is used to evaluate the classification accuracy and generalization ability of the model. By comparing the predicted results of the model on the validation set with the true results, the performance of the model can be evaluated.

[0124] F305: When the model performance reaches the preset standard, save the model parameters. These parameters include the number of decision trees, the node structure of each tree, the splitting conditions, etc. After saving the model parameters, the trained health status assessment model can be obtained.

[0125] The trained health status assessment model can be applied to the actual health status assessment of electromechanical systems. By inputting new feature data, the model can output the health status assessment result, providing decision support for the maintenance and management of the system. At the same time, the model can also be evaluated and updated regularly to adapt to the changes in the system status and new data features.

[0126] Embodiment 3: To realize the remaining life prediction in an electromechanical system, this embodiment adopts a particle filter algorithm to construct a remaining life prediction model. The detailed steps are as follows:

[0127] F401: Data preparation and preprocessing: Select the relevant data of key components from the historical data of the electromechanical system, including maintenance records and operating environment data. Preprocess these data, including data cleaning, format conversion, time synchronization, and outlier detection and processing, to ensure the accuracy and reliability of the data. After the preprocessing is completed, select the historical maintenance records and operating environment data of key components and organize them according to the time series to form a time series dataset.

[0128] F402: Define the state - space model: According to the degradation characteristics of key components and operation environment data, define the state - space model. The state - space model includes a state - transition equation and an observation equation, which are used to describe the dynamic change process of the remaining life of key components and the correlation relationship with environmental data.

[0129] F403: Initialize the particle set: According to the definition of the state - space model, initialize the particle set. Each particle represents a possible value of the remaining life of the key component, and the size of the particle set can be set according to actual needs. When initializing, the initial distribution of particles can be set according to prior knowledge or historical data.

[0130] F404: Recursively update the particle set: According to the state - transition equation and the observation equation, recursively update the particle set. At each time step, use the state - transition equation to predict the state transition of particles, and then use the observation equation to update and correct the state of particles. By continuously recursively updating the particle set, an estimated value of the remaining life of the key component can be obtained.

[0131] F405: Model training and verification: Use the validation set to verify the model. The validation set is a data set independent of the training set, which is used to evaluate the prediction accuracy and stability of the model. By comparing the prediction results of the model on the validation set with the true results, the performance of the model can be evaluated. Evaluation metrics can include prediction error, stability metrics, etc.

[0132] F406: When the model performance reaches the preset standard, save the model parameters. The model parameters include the parameters of the state - space model, the initial distribution and size of the particle set, etc. After saving the model parameters, a trained remaining - life prediction model can be obtained.

[0133] Model application: The trained remaining - life prediction model can be applied to actual electromechanical systems to predict the remaining life of key components in real - time. By inputting new operation environment data and maintenance records, the model can output an estimated value of the remaining life of the key component, providing decision - making support for the maintenance and management of the system.

[0134] Example 4: To achieve the comprehensive health assessment of an electromechanical system, this example details a multi - model fusion algorithm, which can perform fusion analysis on the output results of a fault - diagnosis model, a health - state assessment model, and a remaining - life prediction model, so as to obtain a more accurate and comprehensive comprehensive health assessment result. The specific implementation steps include:

[0135] F601: Define model output

[0136] Mark the output content of each model. The fault type output by the fault diagnosis model is denoted as D_DNN; the health status evaluation result output by the health status evaluation model is denoted as D_RF; the remaining life predicted by the remaining life prediction model is denoted as D_PF.

[0137] F602: Fusion of health status evaluation results

[0138] For D_DNN and D_RF, if the outputs of the two are inconsistent, a weighted average method is used for fusion. The weights of the weighted average can be determined according to the confidence or accuracy of each model. Specifically, based on historical data or performance on the validation set, a weight can be assigned to each model, and then the weighted average is calculated to obtain the fused health status evaluation result D_Fusion_Health.

[0139] F603: Incorporate remaining life prediction

[0140] Combine D_PF output by the remaining life prediction model with the fused health status evaluation result D_Fusion_Health. This can form a comprehensive evaluation result that includes both the current health status and the future remaining life.

[0141] F604: Process multiple key components

[0142] If there are multiple key components in the electromechanical system, the above fusion process needs to be carried out for each component separately. That is, calculate D_DNN, D_RF, and D_PF for each component separately and perform fusion to obtain their respective comprehensive evaluation results.

[0143] F605: Overall comprehensive evaluation

[0144] Based on the comprehensive evaluation results of all key components, use the weighted average method to obtain the final comprehensive health evaluation result D_Final. The weights of the weighted average can be determined according to the importance of each component or the severity of the fault consequences.

[0145] F606: Determine the system health status

[0146] Compare D_Final with a preset health standard or threshold. According to the comparison result, the overall health status, potential faults, and remaining life of the electromechanical system can be determined. For example, different health levels (such as excellent, good, average, poor, dangerous, etc.) can be set, and the system can be classified into the corresponding level according to the value of D_Final.

[0147] The present invention also includes a comprehensive evaluation system for the electromechanical system PHM system, and the system includes:

[0148] A data collection and preprocessing module, which is used to collect the operation data of the electromechanical system, including sensor data, equipment logs, maintenance records, and operating environment data, and preprocess the collected data, including data cleaning, format conversion, time synchronization, and outlier detection, and then perform feature extraction to form a feature dataset;

[0149] A fault diagnosis model construction module, which is used to construct a fault diagnosis model based on the deep neural network DNN algorithm. This model is trained using the sensor data and equipment logs in the feature dataset to classify the fault types of the electromechanical system;

[0150] A health status assessment model construction module, which is used to construct a health status assessment model based on the random forest RF algorithm. This model uses the preprocessed feature dataset to identify the patterns of different health states;

[0151] A remaining useful life prediction model construction module, which is used to construct a remaining useful life prediction model based on the particle filter PF algorithm. This model uses historical maintenance records and operating environment data to predict the remaining useful life of the key components of the electromechanical system;

[0152] A model output acquisition module, which is used to input the feature dataset into the fault diagnosis model, health status assessment model, and remaining useful life prediction model respectively to obtain the fault diagnosis result, health status assessment result, and remaining useful life prediction result of the electromechanical system;

[0153] A multi-model fusion analysis module, which is used to define a multi-model fusion algorithm and perform fusion analysis on the output results obtained by the model output acquisition module to obtain a comprehensive health assessment result;

[0154] An evaluation report generation module, which is used to generate an evaluation report on the PHM system of the electromechanical system according to the comprehensive health assessment result, including fault types, health status, remaining useful life, and corresponding maintenance suggestions.

[0155] The implementation method of this system refers to the above-mentioned embodiments and will not be elaborated in the specification.

[0156] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0157] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A comprehensive evaluation method for a PHM system of a mechatronic system, characterized in that: The method comprises: F1: Collect the operation data of the electromechanical system, including sensor data, equipment logs, maintenance records and operation environment data; pre-process the collected data, including data cleaning, format conversion, time synchronization and outlier detection, and then extract features to form a feature data set; F2: Build a fault diagnosis model based on a deep neural network (DNN) algorithm, which is trained using sensor data and equipment logs in the feature dataset to classify the fault types of the electromechanical system; F3: Build a health status assessment model based on the random forest RF algorithm, which uses the preprocessed feature dataset to learn and identify patterns of different health status; F4: Construct a remaining life prediction model based on the particle filter PF algorithm, which uses historical maintenance records and operating environment data to predict the remaining life of key components of the electromechanical system; F5: Input the characteristic data set into the fault diagnosis model, health status assessment model and remaining life prediction model respectively to obtain the fault diagnosis results, health status assessment results and remaining life prediction results of the electromechanical system; F6: Define the multi-model fusion algorithm, perform fusion analysis on the model output results, and obtain comprehensive health assessment results; F7: Generate a PHM system assessment report for the electromechanical system based on the comprehensive health assessment results, including fault type, health status, remaining life and corresponding maintenance recommendations.

2. A comprehensive evaluation method for a PHM system of an electromechanical system according to claim 1, characterized in that: The sensor data includes temperature, pressure, vibration and current data; the equipment log includes operating time, number of shutdowns and fault history; the maintenance record includes maintenance time, maintenance type and maintenance cost; the operating environment data includes ambient temperature, humidity and dust concentration.

3. A comprehensive evaluation method for a PHM system of an electromechanical system according to claim 1, characterized in that: The fault diagnosis model adopts a multi-layer perceptron MLP structure, extracts features through multiple fully connected layers, and finally outputs the fault type through a Softmax layer. Its network structure includes: Input layer: receives preprocessed sensor data and device logs, with the data dimension unified as [N, F], where N is the number of samples and F is the number of features; Hidden layer: multi-layer fully connected operation is adopted, and each layer of full connection is followed by ReLU activation function to extract features; Output layer: The Softmax function is used to output the probability of each fault type, and the number of output nodes is the number of fault types.

4. A comprehensive evaluation method for a PHM system of a mechatronic system according to claim 3, characterized in that: The training steps of the fault diagnosis model include: F201: Initialize the parameters of the DNN model, including weights, bias terms, and learning rates; F202: Divide the preprocessed feature data set into a training set and a validation set. The training set is used for model training, and the validation set is used for model validation. F203: Use the training set to iteratively train the DNN model. In each iteration, the model output is calculated by forward propagation, and the loss between the predicted value and the true value is calculated according to the loss function; the loss function adopts the cross entropy loss function; F204: Update the model parameters by back propagation algorithm and gradient descent method according to the loss value; the formula for updating the model parameters by gradient descent method is: in, Represents the model parameters, including all weights and biases that the model needs to learn; Represents the learning rate, which is used to control the step size of parameter update; Represents the loss function About parameters The gradient of is a vector pointing to the direction where the loss function grows fastest; Indicates an assignment operation, used to update parameters The value of F205: Repeat steps F103 to F104 until the loss converges or the preset number of iterations is reached; F206: Use the validation set to validate the trained DNN model and evaluate the classification accuracy of the model; F207: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained fault diagnosis model.

5. The comprehensive evaluation method of the electromechanical system PHM system according to claim 1 is characterized in that: The health status assessment model uses multiple decision trees to construct a random forest and outputs the health status assessment results through a voting mechanism. Its structure includes: Input layer: receives the preprocessed feature data set; Decision tree layer: build multiple decision trees, each tree independently classifies the input data; Output layer: A voting mechanism is used to integrate the classification results of all decision trees and output the health status assessment results.

6. A comprehensive evaluation method for a PHM system of a mechatronic system according to claim 5, characterized in that: The steps to train the health status assessment model include: F301: Use the preprocessed feature dataset as the training set and select health status as the label; F302: Determine the number and maximum depth of decision trees in a random forest; F303: Use the training set to train the random forest model until the model converges; F304: Use the validation set to validate the random forest model and evaluate the classification accuracy and generalization ability of the model; F305: When the model performance reaches the preset standard, the model parameters are saved to obtain a trained health status assessment model.

7. The comprehensive evaluation method of the electromechanical system PHM system according to claim 1 is characterized in that: The remaining life prediction model adopts a particle filter algorithm to estimate the remaining life of key components by recursion, and its implementation steps include: F401: Filter out the historical maintenance records and operating environment data of key components from the preprocessed feature data set to form a time series data set; F402: Define the state space model, including the state transfer equation and observation equation; F403: Initialize the particle set, each particle represents a possible value of the remaining life of the key component; F404: According to the state transfer equation and observation equation, recursively update the particle set to obtain the estimated value of the remaining life; F405: Use the validation set to validate the model and evaluate the prediction accuracy and stability of the model; F406: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained remaining life prediction model.

8. The comprehensive evaluation method of the electromechanical system PHM system according to claim 1 is characterized in that: The specific implementation steps of defining the multi-model fusion algorithm in step F6 include: F601: Define the fault type output by the fault diagnosis model as D_DNN, the health status assessment result output by the health status assessment model as D_RF, and the remaining life output by the remaining life prediction model as D_PF; F602: For D_DNN and D_RF, if the output results of the two are inconsistent, a weighted average is performed according to the confidence or accuracy of each model to obtain the fused health status assessment result D_Fusion_Health; F603: Combine D_PF with the fused health status assessment result D_Fusion_Health to form a comprehensive health assessment result that includes the current health status and future remaining life span; F604: If there are multiple key components, the above fusion process is performed on each component separately to obtain their respective comprehensive evaluation results; F605: Based on the comprehensive evaluation results of all key components, the weighted average method is used to obtain the final comprehensive health evaluation result D_Final; F606: Compare D_Final with preset health standards or thresholds to determine the overall health status, potential failures and remaining life of the electromechanical system.

9. The comprehensive evaluation method of the electromechanical system PHM system according to claim 1 is characterized by: In the evaluation report generated in step S7, the comprehensive evaluation results are visualized, including drawing a health status trend graph, a remaining life distribution graph, and a fault type ratio graph.

10. A comprehensive evaluation system for the PHM system of an electromechanical system, characterized in that: The system comprises: The data collection and preprocessing module is used to collect the operating data of the electromechanical system, including sensor data, equipment logs, maintenance records and operating environment data, and preprocess the collected data, including data cleaning, format conversion, time synchronization and outlier detection, and then extract features to form a feature data set; A fault diagnosis model building module is used to build a fault diagnosis model based on a deep neural network (DNN) algorithm. The model is trained using sensor data and equipment logs in the feature data set to classify the fault type of the electromechanical system. The health status assessment model building module is used to build a health status assessment model based on the random forest RF algorithm. The model uses the preprocessed feature data set to identify patterns of different health states; The remaining life prediction model building module is used to build a remaining life prediction model based on the particle filter PF algorithm. The model uses historical maintenance records and operating environment data to predict the remaining life of key components of the electromechanical system; The model output acquisition module is used to input the characteristic data set into the fault diagnosis model, the health status assessment model and the remaining life prediction model respectively to obtain the fault diagnosis result, the health status assessment result and the remaining life prediction result of the electromechanical system; The multi-model fusion analysis module is used to define the multi-model fusion algorithm, perform fusion analysis on the output results obtained by the model output acquisition module, and obtain comprehensive health assessment results; The assessment report generation module is used to generate a PHM system assessment report for the electromechanical system based on the comprehensive health assessment results, including fault type, health status, remaining life and corresponding maintenance recommendations.

Images