Equipment health assessment method based on artificial intelligence and electronic equipment

Abstract

The invention discloses an equipment health assessment method based on artificial intelligence and electronic equipment, and belongs to the technical field of equipment assessment. The method comprises the steps that firstly, heterogeneous data such as sensor data, historical maintenance records and environmental parameters in the equipment operation process are acquired through a multi-source data acquisition system, and performance index data of equipment manufacturers and user feedback information are integrated; then, a multi-step preprocessing process including data cleaning, feature extraction, normalization processing, feature dimension reduction based on principal component analysis and the like is adopted, and the problems of data redundancy and noise interference are effectively solved; a deep learning model based on a convolutional neural network or a recurrent neural network is constructed, and a dynamic weight adjustment mechanism is introduced, so that the model can adapt to running state changes of different devices; and finally, outputting an evaluation report containing health grade division and future state prediction, and generating corresponding early warning information and maintenance suggestions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment evaluation, in particular to an equipment health evaluation method based on artificial intelligence and an electronic device. BACKGROUND

[0002] With the rapid advancement of industrial intelligence, equipment health evaluation technology, as a key link in the intelligent manufacturing system, is undergoing a fundamental change from traditional experience judgment to intelligent decision-making.

[0003] In modern industrial production scenarios, equipment health evaluation not only concerns the stable operation of a single device, but also is a core element affecting the reliability, safety and economy of the entire production system. According to statistical data, effective implementation of equipment health evaluation can reduce unplanned downtime by more than 40%, reduce maintenance costs by 25%-30%, and improve overall equipment effectiveness (OEE) by 15%-20%. These significant benefits directly promote the rapid popularization of the technology in global manufacturing; traditional evaluation methods mainly follow two technical routes: one is qualitative evaluation based on human experience, which relies on technical personnel to make subjective judgments through "observation, smell, questioning and palpation" based on sensory indicators such as equipment abnormal sound, abnormal vibration and temperature change. This method, although effective in certain specific scenarios, is severely limited by personnel experience levels and is difficult to achieve standardized and large-scale applications; the other is quantitative evaluation based on simple data analysis, which uses basic signal processing methods such as time domain statistics (such as effective value, peak factor), frequency domain analysis (such as FFT transform). This method, although improving the objectivity of evaluation, has obvious limitations in dealing with high-dimensional, nonlinear and non-stationary signals generated by modern industrial equipment, especially in complex operating environments such as variable working conditions and strong interference, the evaluation accuracy often decreases significantly.

[0004] In recent years, breakthroughs in artificial intelligence technology have revolutionized equipment health assessment. The new generation of algorithms represented by deep learning has shown strong feature extraction and pattern recognition capabilities. In terms of specific applications, evaluation methods based on machine learning classification algorithms have become the industry mainstream. For example, support vector machines (SVM) can achieve 85%-92% accuracy in rotating machinery fault classification, and random forest algorithms show excellent anti-interference performance when processing multi-sensor fusion data. More advanced deep learning methods such as convolutional neural networks (CNN) can automatically learn the time-frequency features of vibration signals and achieve over 95% recognition accuracy in bearing fault diagnosis. Long short-term memory networks (LSTM) are good at capturing the time sequence degradation rules of equipment performance and have been successfully applied to the residual life prediction of aircraft engines. The introduction of knowledge graph technology has created a new paradigm for multi-source information fusion. By constructing a knowledge graph containing device structure, fault mode, maintenance history, and other elements, and using graph neural networks (GNN) for multi-set pooling operations, cross-modal data correlation reasoning is achieved. After applying this technology to a large petrochemical enterprise, the health assessment accuracy of key units was improved by 18 percentage points.

[0005] However, existing methods still face several technical bottlenecks that need to be addressed. In terms of multi-source data fusion, the vibration signals, process parameters, infrared images, and acoustic emission data generated by modern industrial equipment are significantly heterogeneous. Traditional feature-level splicing or decision-level weighting methods cannot effectively extract deep correlation information. For example, in a wind turbine gearbox case, relying solely on vibration signals missed 15% of early-stage lubrication faults, and simple fusion of multi-source data led to an explosion in feature dimensionality. In terms of dynamic weight adjustment, existing evaluation models often use fixed weight coefficients, which cannot adapt to the dynamic process of equipment performance degradation. A study on a numerical control machine tool showed that after 8000 hours of operation, the evaluation error of a traditional static weight model was as high as 32%. In terms of complex system adaptability, for the electro-hydraulic coupling effects in large-scale complete equipment (such as automobile welding lines and steel rolling production lines), the existing "divide and conquer" evaluation strategy cannot establish a system-level health state mapping, resulting in about 30% of related faults being misjudged as independent events. These technical bottlenecks seriously restrict the practical application effect of health assessment systems in industrial sites and need to be broken through through multidisciplinary cross-innovation, including developing adaptive fusion architectures based on attention mechanisms, constructing dynamic evaluation frameworks with online learning capabilities, and using digital twin technology to realize the organic fusion of physical models and data-driven methods. In the future, with the maturation of emerging technologies such as 5G communication, edge computing, and quantum sensing, equipment health assessment will develop towards real-time, precision, and systematization, ultimately forming a closed-loop intelligent operation and maintenance system of "perception- assessment-decision-optimization" to provide solid technical support for the development of industrial intelligence. SUMMARY

[0006] The present application provides an equipment health assessment method based on artificial intelligence and an electronic device, which can overcome some or some defects of the prior art.

[0007] According to the equipment health assessment method based on artificial intelligence, the method comprises the following steps:

[0008] S1, collecting data during equipment operation: using a distributed data collection system, through a multi-type sensor array of vibration sensors, temperature sensors, current sensors and the like installed at key parts of the equipment, real-time collection of physical quantity data such as vibration signals, temperature distribution, energy consumption parameters and the like during equipment operation, the sampling frequency is set to 1kHz-10kHz according to the characteristics of the equipment; at the same time, the historical maintenance records of the equipment are obtained through the industrial Internet of Things platform, including maintenance time, replacement parts, fault description and the like; and the equipment operating environment parameters are collected, including environmental temperature, humidity, dust concentration and the like. The data collection process uses time synchronization technology to ensure the time sequence consistency of multi-source data;

[0009] S2, pre-processing the collected data to obtain pre-processed data: establishing a data quality control system, detecting and repairing outliers of the collected original data, and eliminating signal noise by using a sliding window method; extracting features such as peak value and effective value through time domain analysis, and extracting frequency spectrum features through frequency domain analysis; text mining is performed on unstructured maintenance records to extract key fault information; finally, all feature data are time-aligned to form a unified format feature matrix;

[0010] S3, constructing a health assessment model based on deep learning, and training the model using the pre-processed data to obtain a trained health assessment model: designing a hybrid neural network architecture, using a 3-layer convolutional neural network in the front end to extract local spatio-temporal features, and using a bidirectional LSTM network in the back end to capture time sequence dependency; the input layer of the model is designed as a multi-channel structure, which respectively receives feature data of different modalities; an attention mechanism is used to realize adaptive weighting of features; a transfer learning strategy is used in the training process, which is pre-trained on a large-scale simulation data, and then fine-tuned with actual equipment data;

[0011] S4, inputting the real-time collected equipment operation data into the trained health assessment model to obtain an equipment health assessment report: the model output includes quantitative indexes such as health score (0-100 points), fault probability (0-1), and remaining service life prediction (hours); at the same time, visual analysis charts are generated to show the change trend of key parameters and abnormal features; the evaluation result is updated every 10 minutes, supporting real-time monitoring and historical backtracking;

[0012] S5, generating early warning information and maintenance recommendations for the equipment based on the evaluation report: setting a three-level early warning mechanism (attention, warning, danger) according to the health score, and automatically triggering alarm signals of different levels; the maintenance recommendation system combines failure probability, historical maintenance records and equipment manuals to recommend the optimal maintenance plan, including maintenance timing, required spare parts and estimated time consumption.

[0013] As preferred, the real-time collected equipment operation data includes sensor data: three-dimensional vibration acceleration signals (sampling rate ≥ 5 kHz), temperature distribution data (accuracy ± 0.5℃), motor current waveform (sampling rate 1 kHz), oil particle count, etc.

[0014] Historical maintenance records: structured database records of repair work orders, unstructured repair report texts, serial numbers and service life of replaced parts, etc.

[0015] And environmental parameters: real-time temperature and humidity (measurement range -20℃-60℃, accuracy ±2%RH), dust concentration (detection range 0-20mg / m 3 ), corrosive gas content, etc. obtained through environmental monitoring terminals.

[0016] As preferred, the pre-processed data includes equipment manufacturer's performance indicator data: rated parameters in design specifications, factory test reports, performance decay curves within the warranty period, etc.

[0017] User feedback information: non-normal operating conditions recorded in the operation log, suspicious phenomena found by manual inspection, and unstructured data such as user subjective evaluation.

[0018] As preferred, the pre-processing of data includes the following steps:

[0019] S21, data cleaning and feature extraction are performed on the collected data;

[0020] Among them, data cleaning: using 3σ rule-based outlier detection, linear interpolation repair for data points exceeding ±3 times the standard deviation; for missing data, use weighted average of adjacent time points to fill in;

[0021] Feature extraction: 12-dimensional statistical features (including peak value, peak-to-peak value, waveform index, etc.) are extracted in time domain, 20-dimensional frequency spectrum features are extracted by FFT transformation in frequency domain, and 15-dimensional energy features are extracted by wavelet transformation in time-frequency domain;

[0022] S22, normalizing the extracted data features; using the maximum and minimum value normalization method to map all features to the [0, 1] interval; for category type features, use one-hot encoding;

[0023] S23, based on principal component analysis method to the normalized data characteristics of feature dimension reduction; calculate the feature correlation coefficient matrix, and retain the cumulative contribution rate of 95% of the principal component; finally, the original 47-dimensional feature is reduced to 15-dimensional, and the correlation between features is less than 0.3.

[0024] As preferred, the deep learning model adopts one of the following architectures:

[0025] Convolutional neural network variant: contains 4 convolutional layers (convolution kernel size 3x3, step 1), each layer followed by batch normalization and ReLU activation, followed by global average pooling and 3 layers of fully connected;

[0026] Recurrent neural network variant: bidirectional GRU network, hidden layer unit number 128, time step 60, output layer connected with self-attention mechanism;

[0027] Hybrid architecture: the first 3 layers of CNN extract spatial features, followed by 2 layers of BiLSTM to process temporal relationships, and finally through an attention layer for fusion.

[0028] As preferred, when training the health assessment model, a dynamic weight adjustment mechanism is adopted: a weight decay function based on equipment running time is designed, and the weight of recent data increases exponentially with time; automatically adjust the feature weight for different working conditions, such as increasing the temperature feature weight by 30% in high temperature environment; through online learning module, automatically adjust the model parameters once every 24 hours; adapt to the running state of different devices.

[0029] As preferred, when the performance indicator data of equipment manufacturers and user feedback information change, the health assessment model is updated in real time; the model updating mechanism includes:

[0030] When the equipment manufacturer releases new performance indicators, trigger incremental training of the model;

[0031] When the user feedback information accumulates to a certain threshold (such as 100), start model optimization;

[0032] Elastic weight solidification technology is adopted to prevent new data from covering old knowledge.

[0033] As preferred, the health assessment report of the equipment includes the health grade division of the equipment; the health assessment report of the equipment includes:

[0034] Health grade division: divide the equipment state into 5 levels (excellent ≥90 points, good 80-89 points, attention 70-79 points, warning 60-69 points, danger <60 points);

[0035] Key component health radar chart, intuitively display the state of each subsystem;

[0036] Health trend curve, showing the score change in the last 30 days.

[0037] As preferred, the health assessment report of the equipment further comprises:

[0038] LSTM-based remaining useful life prediction, confidence interval display;

[0039] Failure mode probability distribution chart;

[0040] Maintenance urgency index (1-10 points);

[0041] Similar failure case recommendation.

[0042] As preferred, the health assessment report of the equipment further comprises a prediction of the future health state of the equipment.

[0043] An electronic device comprises:

[0044] At least one processor; an industrial-grade processor with a multi-core architecture, a main frequency of 2.4GHz or more, and an AI acceleration module;

[0045] A memory in communication connection with the at least one processor; including 16GB or more of DDR4 memory and 1TB or more of SSD storage;

[0046] Data acquisition interface: supporting industrial protocols such as Modbus and OPC UA;

[0047] Communication module: dual-gigabit Ethernet, optional 5G module;

[0048] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

[0049] The present application realizes accurate assessment of the health state of the equipment by fusing multi-source data and deep learning technology, more comprehensively reflects the actual running state of the equipment, automatically extracts complex features in the data by using a deep learning model, improves the accuracy and efficiency of the assessment, significantly reduces the maintenance cost of the equipment, improves the running efficiency and reliability of the equipment, and has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of the equipment health assessment method based on artificial intelligence of the present embodiment;

[0051] Figure 2 A flowchart of the data preprocessing step of the present embodiment. DETAILED DESCRIPTION

[0052] For further understanding of the present application, the present application will be described in detail in conjunction with the embodiments. It should be understood that the embodiments are only to explain the present application but not to limit the present application.

[0053] Embodiment 1

[0054] The embodiment provides an equipment health evaluation method based on artificial intelligence, which can overcome some or some defects of the prior art.

[0055] As Figure 1 shown, the equipment health evaluation method based on artificial intelligence according to the embodiment comprises the following steps:

[0056] S1, collecting data in the running process of equipment: adopting a distributed data collection system, through a multi-type sensor array of vibration sensors, temperature sensors, current sensors and the like installed at key parts of the equipment, real-time collection of physical quantity data such as vibration signals, temperature distribution, energy consumption parameters and the like when the equipment is running, the sampling frequency is set to 1 kHz-10 kHz according to the characteristics of the equipment; at the same time, the historical maintenance records of the equipment are obtained through an industrial Internet of Things platform, including maintenance time, replacement parts, fault description and the like; and the equipment operating environment parameters are collected, including environmental temperature, humidity, dust concentration and the like. The data collection process adopts time synchronization technology to ensure the time sequence consistency of multi-source data;

[0057] S2, pre-processing the collected data to obtain pre-processed data: establishing a data quality control system, detecting and repairing outliers of the collected original data, and eliminating signal noise by using a sliding window method; extracting features such as peak value and effective value through time domain analysis, and extracting frequency spectrum features through frequency domain analysis; text mining is performed on unstructured maintenance records to extract key fault information; finally, all feature data are time-aligned to form a unified format feature matrix;

[0058] S3, constructing a health evaluation model based on deep learning, and training the model using the pre-processed data to obtain a trained health evaluation model: designing a hybrid neural network architecture, using a 3-layer convolutional neural network in the front end to extract local space-time features, and using a bidirectional LSTM network in the back end to capture time sequence dependency; the input layer of the model is designed as a multi-channel structure, respectively receiving feature data of different modalities; an attention mechanism is used to realize adaptive weighting of features; a transfer learning strategy is used in the training process, pre-training on a large-scale simulation data, and then fine-tuning on actual equipment data;

[0059] S4. Input the real-time collected equipment operation data into the trained health assessment model to obtain the equipment health assessment report: the model output includes quantitative indicators such as health score (0-100 points), failure probability (0-1), and remaining service life prediction (hours). At the same time, a visual analysis chart is generated to show the trend of key parameter changes and abnormal characteristics. The assessment results are updated every 10 minutes, supporting real-time monitoring and historical backtracking.

[0060] S5. Generate early warning information and maintenance recommendations for equipment based on the assessment report: Set a three-level early warning mechanism (attention, warning, danger) based on the health score, and automatically trigger alarm signals of different levels; the maintenance recommendation system combines the failure probability, historical maintenance records and equipment manuals to recommend the optimal maintenance plan, including maintenance timing, required spare parts and estimated time.

[0061] Among them, the equipment operation data collected in real time includes sensor data: three-dimensional vibration acceleration signal (sampling rate ≥ 5kHz), temperature distribution data (accuracy ±0.5°C), motor current waveform (sampling rate 1kHz), oil particle count, etc.

[0062] Historical maintenance records: maintenance work orders recorded in a structured database, unstructured maintenance report texts, serial numbers and service lives of replaced parts, etc.

[0063] And environmental parameters: real-time temperature and humidity (measurement range -20℃~60℃, accuracy ±2%RH), dust concentration (detection range 0~20mg / m 3 ), corrosive gas content, etc.;

[0064] By integrating this data, the operating status of the equipment can be more comprehensively reflected, providing richer information for equipment health assessment.

[0065] The pre-processed data includes the equipment manufacturer's performance indicator data: rated parameters in the design specification, factory test reports, performance degradation curves during the warranty period, etc.

[0066] User feedback information: unstructured data such as abnormal operating conditions recorded in operation logs, suspicious phenomena discovered during manual inspections, and user subjective evaluations.

[0067] Among them, such as Figure 2 As shown, the data preprocessing includes the following steps:

[0068] S21, cleaning the collected data and performing feature extraction;

[0069] Among them, data cleaning: adopt 3σ criterion-based outlier detection, linear interpolation repair for data points exceeding ±3 times the standard deviation; for missing data, adopt adjacent time point weighted average filling;

[0070] Feature extraction: 12-dimensional statistical features (including peak value, peak-to-peak value, waveform index, etc.) are extracted in the time domain; 20-dimensional frequency spectrum features are extracted by FFT transformation in the frequency domain; 15-dimensional energy features are extracted by wavelet transformation in the time-frequency domain;

[0071] S22, normalize the extracted data features; adopt the maximum and minimum value normalization method to map all features to the [0, 1] interval; adopt one-hot encoding for category type features;

[0072] S23, perform feature dimension reduction on the normalized data features based on principal component analysis; calculate the feature correlation coefficient matrix and retain the principal components with cumulative contribution rate ≥95%; finally, reduce the original 47-dimensional features to 15-dimensional features with correlation between features <0.3.

[0073] Feature dimension reduction of data features can reduce data redundancy and improve model training efficiency.

[0074] Among them, the deep learning model adopts one of the following architectures:

[0075] Convolutional neural network variant: contains 4 convolutional layers (convolution kernel size 3×3, step 1), each layer is followed by batch normalization and ReLU activation, and finally connected to global average pooling and 3 layers of full connection;

[0076] Recurrent neural network variant: bidirectional GRU network, hidden layer unit number 128, time step 60, output layer connected to self-attention mechanism;

[0077] Hybrid architecture: the first 3 layers of CNN extract spatial features, followed by 2 layers of BiLSTM to process temporal relationships, and finally fused through an attention layer.

[0078] Among them, when training the health assessment model, a dynamic weight adjustment mechanism is adopted: a weight decay function based on equipment running time is designed, and the weight of recent data increases exponentially over time; automatically adjust the feature weight for different working conditions, such as increasing the temperature feature weight by 30% in high temperature environment; through the online learning module, the model parameters are automatically adjusted once every 24 hours; adapt to the running state of different devices.

[0079] Among them, when the performance index data of equipment manufacturers and user feedback information change, the health assessment model is updated in real time; the model update mechanism includes:

[0080] When the equipment manufacturer releases new performance indicators, trigger incremental training of the model;

[0081] User feedback information accumulates to a certain threshold (such as 100) to start model optimization;

[0082] Adopt elastic weight solidification technology to prevent new data from covering old knowledge.

[0083] Among them, the health assessment report of the equipment includes the health level division of the equipment; the health assessment report of the equipment includes:

[0084] Health level division: divide the equipment state into 5 levels (excellent ≥90 points, good 80-89 points, attention 70-79 points, warning 60-69 points, danger <60 points);

[0085] Key component health radar chart, intuitively display the state of each subsystem;

[0086] Health trend curve, showing the score change in the last 30 days.

[0087] Among them, the health assessment report of the equipment also includes:

[0088] LSTM-based remaining useful life prediction with confidence interval display;

[0089] Fault mode probability distribution chart;

[0090] Maintenance urgency index (1-10 points);

[0091] Similar fault case recommendation.

[0092] The embodiment realizes accurate assessment of the health state of the equipment by fusing multi-source data and deep learning technology, more comprehensively reflects the actual running state of the equipment, automatically extracts complex features in the data by using a deep learning model, improves the accuracy and efficiency of the assessment, significantly reduces the maintenance cost of the equipment, improves the running efficiency and reliability of the equipment, and has a wide application prospect.

[0093] Embodiment 2

[0094] The embodiment provides an electronic device which can overcome some or some defects of the prior art.

[0095] The electronic device according to the embodiment comprises:

[0096] At least one processor; an industrial-grade processor with a multi-core architecture, a main frequency ≥2.4GHz, and an AI acceleration module; and

[0097] The memory in communication connection with the at least one processor; including 16GB above DDR4 memory, 1TB above SSD storage;

[0098] Data acquisition interface: support Modbus, OPC UA and other industrial protocols;

[0099] Communication module: dual gigabit Ethernet, optional 5G module;

[0100] The memory stores instructions executable by the at least one processor, the instructions executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

[0101] It is easy to understand that, on the basis of one or more embodiments provided in the present application, other embodiments can be obtained by combining, splitting, recombining, etc. of the embodiments of the present application, and these embodiments do not exceed the protection scope of the present application.

[0102] The above describes the present application and its embodiments in a schematic manner, and the description is not restrictive, and the embodiments shown are only part of the embodiments of the present application, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired thereby, without departing from the spirit of the present application, similar structural modes and embodiments can be designed without creativity, and all should belong to the protection scope of the present application.

Claims

1. An equipment health assessment method based on artificial intelligence, characterized in that: The following steps are involved: S1. Data collection during equipment operation: A distributed data acquisition system is used to collect real-time data on physical quantities such as vibration signals, temperature distribution, and energy consumption parameters during equipment operation through a multi-type sensor array installed at key locations of the equipment, including vibration sensors, temperature sensors, and current sensors. The sampling frequency is set to 1kHz-10kHz based on the characteristics of the equipment. The system also uses the Industrial Internet of Things platform to obtain historical maintenance records of the equipment, including repair time, replacement parts, and fault descriptions. Furthermore, the system collects parameters related to the equipment's operating environment, including ambient temperature, humidity, and dust concentration. Time synchronization technology is used during the data collection process to ensure temporal consistency of multi-source data. S2. Preprocess the collected data to obtain preprocessed data: establish a data quality control system, detect and repair outliers on the collected raw data, and use a sliding window method to eliminate signal noise; extract features such as peak values ​​and effective values ​​through time domain analysis, and extract spectral features through frequency domain analysis; Perform text mining on unstructured maintenance records to extract key fault information; finally, time-align all feature data to form a feature matrix in a unified format; S3. Build a deep learning-based health assessment model and train it using preprocessed data to obtain a trained health assessment model: Design a hybrid neural network architecture, using a three-layer convolutional neural network on the front end to extract local spatiotemporal features and a bidirectional LSTM network on the back end to capture temporal dependencies. The model input layer is designed as a multi-channel structure to receive feature data from different modalities. Use an attention mechanism to achieve adaptive feature weighting. The training process adopts a transfer learning strategy, first pre-training on large-scale simulation data, and then fine-tuning using actual equipment data; S4. Input the real-time collected equipment operation data into the trained health assessment model to obtain the equipment health assessment report. The model outputs quantitative indicators including health score, failure probability, and remaining service life prediction. It also generates visual analysis charts to show the trend of key parameter changes and abnormal characteristics. The assessment results are updated every 10 minutes, supporting real-time monitoring and historical backtracking. S5. Generate early warning information and maintenance recommendations for equipment based on the assessment report: Set a three-level early warning mechanism of attention, warning, and danger based on the health score, and automatically trigger alarm signals of different levels; the maintenance recommendation system combines the failure probability, historical maintenance records and equipment manuals to recommend the optimal maintenance plan, including maintenance timing, required spare parts and estimated time.

2. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: The equipment operation data collected in real time includes sensor data: three-dimensional vibration acceleration signals, temperature distribution data, motor current waveforms, and oil particle counts; Historical maintenance records: maintenance work orders recorded in a structured database, unstructured maintenance report texts, serial numbers and service life of replaced parts; And environmental parameters: real-time temperature and humidity, dust concentration, and corrosive gas content obtained through the environmental monitoring terminal.

3. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: The data to be pre-processed includes the equipment manufacturer's performance indicator data: rated parameters in the design specification, factory test reports, and performance degradation curves during the warranty period; User feedback information: unstructured data such as abnormal operating conditions recorded in operation logs, suspicious phenomena discovered during manual inspections, and user subjective evaluations.

4. The equipment health assessment method based on artificial intelligence according to claim 3 is characterized in that: The data preprocessing includes the following steps: S21, cleaning the collected data and performing feature extraction; Among them, data cleaning: outlier detection based on the 3σ criterion is used, and linear interpolation is performed on data points exceeding ±3 times the standard deviation; missing data are filled by weighted average of adjacent time points; Feature extraction: 12-dimensional statistical features are extracted in the time domain, 20-dimensional spectrum features are extracted in the frequency domain through FFT transformation, and 15-dimensional energy features are extracted in the time-frequency domain through wavelet transformation; S22. Normalize the extracted data features; use the maximum and minimum normalization method to map all features to the [0, 1] interval; use one-hot encoding for categorical features; S23. Perform feature dimensionality reduction on the normalized data features based on principal component analysis. Calculate the feature correlation coefficient matrix and retain the principal components with a cumulative contribution rate ≥ 95%. Finally, reduce the original 47-dimensional features to 15 dimensions, with inter-feature correlation < 0.

3.

5. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: Deep learning models use one of the following architectures: Convolutional neural network variant: contains 4 convolutional layers, each followed by batch normalization and ReLU activation, and finally followed by global average pooling and 3 layers of full connection; Recurrent Neural Network variant: bidirectional GRU network with 128 hidden units, 60 time steps, and self-attention mechanism at the output layer; Hybrid architecture: The first three layers of CNN extract spatial features, followed by two layers of BiLSTM to process temporal relationships, and finally fused through the attention layer.

6. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: When training the health assessment model, a dynamic weight adjustment mechanism is adopted: a weight decay function based on the equipment's operating time is designed, and the weight of recent data increases exponentially with time; feature weights are automatically adjusted for different working conditions, such as increasing the weight of temperature features by 30% in high-temperature environments; and model parameters are automatically adjusted every 24 hours through an online learning module to adapt to the operating status of different devices.

7. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: When the performance indicator data of equipment manufacturers and user feedback information changes, the health assessment model is updated in real time; The model update mechanism includes: When equipment manufacturers release new performance indicators, incremental model training is triggered; When user feedback information accumulates to a certain threshold, model optimization is initiated; Adopt elastic weight solidification technology to prevent new data from overwriting old knowledge.

8. The equipment health assessment method based on artificial intelligence according to claim 1, characterized in that: The equipment health assessment report includes the equipment health level classification; The equipment health assessment report includes: Health level division: equipment status is divided into 5 levels; Key component health radar chart, intuitively showing the status of each subsystem; The health trend curve shows the score changes in the last 30 days.

9. The equipment health assessment method based on artificial intelligence according to claim 8, characterized in that: The equipment health assessment report also includes: LSTM-based remaining useful life prediction with confidence interval display; Failure mode probability distribution diagram; Maintenance Urgency Index; Recommended similar fault cases.

10. An electronic device comprising: at least one processor; Industrial-grade processor with multi-core architecture, main frequency ≥ 2.4GHz, equipped with AI acceleration module; as well as Memory communicatively connected to the at least one processor; including more than 16GB of DDR4 memory and more than 1TB of SSD storage; Data acquisition interface: supports Modbus, OPC UA and other industrial protocols; Communication module: Dual Gigabit Ethernet, optional 5G module; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

Images