Equipment health management system and method based on multi-channel parameter diagram

Abstract

The invention discloses an equipment health management system and method based on a multi-channel parameter diagram. A parameter acquisition unit acquires health parameters of equipment; the parameter diagram construction unit is used for converting the collected equipment health parameters into equipment parameter diagram data and constructing an equipment parameter diagram; the health condition evaluation unit obtains an equipment health level according to the equipment parameter diagram; the fault level evaluation unit establishes a unified fault level, and obtains an equipment fault level according to the equipment health level; and the maintenance management unit performs attribution analysis according to the equipment fault level to obtain an equipment maintenance scheme. According to the method, firstly, through a unified data architecture and a cross-industry fault system, rapid access of multiple types of equipment is supported, and evaluation management is carried out; and secondly, maintenance suggestions are generated driven by a large language model, automatic closed loop from fault detection to a maintenance scheme is realized, and scientific basis and intelligent suggestions are provided for equipment maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment health management, and more specifically, to an equipment health management system and method based on a multi-channel parameter graph. Background Art

[0002] An operational system relies on a wide variety of equipment, and the health of each piece of equipment determines the success of the system's mission. Accurately and promptly understanding the health of these equipment during operation is crucial to ensuring proper operation and improving production efficiency and safety.

[0003] However, existing equipment health management systems face challenges such as a wide variety of equipment types, incomplete data collection, inaccurate health assessments, and inconsistent fault management standards across industries. Traditional data processing and assessment methods struggle to efficiently adapt to the monitoring and assessment requirements of diverse equipment types, and are unable to accurately identify equipment anomalies and locate the root causes of faults, resulting in a lack of targeted and effective maintenance.

[0004] Currently, various industries (such as manufacturing, aerospace, automotive, energy, and healthcare) have specific fault classification standards, but these indicators and impact levels may vary. Therefore, establishing a unified fault classification system across different industries and equipment types, with clear divisions in impact levels, to enable managers to identify problems across different sectors and make quick decisions, is a challenge.

[0005] Therefore, the problems existing in the prior art need to be further improved and developed. Summary of the Invention

[0006] (1) Purpose of the invention: In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide an equipment health management system and method based on a multi-channel parameter diagram to solve the equipment health management problems existing in the prior art, realize accurate assessment of equipment health status, unified management of fault levels, and intelligent suggestions for repair and maintenance.

[0007] (II) Technical Solution: To solve the above technical problems, this technical solution provides an equipment health management system based on a multi-channel parameter map, including a parameter acquisition unit, a parameter map construction unit, a health status assessment unit, a fault level assessment unit, and a maintenance management unit; the parameter acquisition unit collects health parameters of the equipment;

[0008] The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map;

[0009] The health status assessment unit obtains the equipment health level based on the equipment parameter map;

[0010] The fault level assessment unit establishes a unified fault level and obtains the equipment fault level based on the equipment health level;

[0011] The maintenance management unit conducts attribution analysis based on the equipment failure level and obtains the equipment maintenance plan.

[0012] Preferably, the equipment health parameters include dynamic parameters and static parameters. The dynamic parameters are mainly time-series parameters acquired by sensors in real time during the operation of the equipment. Data loss during data collection can lead to inference deviations. Measures must be taken to address the problem of incomplete data, including:

[0013] Step 101: Perform integrity check on the collected equipment health parameter data;

[0014] Step 102: interpolate the missing data.

[0015] Step 103: predict and issue an alarm for abnormal values;

[0016] Step 104: Verify and optimize the completed parameter data.

[0017] Preferably, the parameter graph construction unit converts the time-series equipment health parameters into time series data according to a certain time length: equipment parameter graph data, and then the time series data is converted into the equipment parameter graph form to spatialize the time series data.

[0018] Preferably, the parameter map construction unit specifically comprises: step 201, performing data preprocessing on equipment health parameters to obtain normalized parameters;

[0019] Step 202: Divide the normalized parameters according to the time series to obtain equipment parameter graph data;

[0020] Step 203: normalize the equipment parameter map data to obtain normalized image data;

[0021] Step 204: Combine the normalized image data into a two-dimensional equipment parameter map.

[0022] Preferably, the time series of dynamic parameters is divided according to windows of specified length, different parameters in each time window are used as image data of different channels, the data of all channels are normalized to obtain normalized image data, the normalized image data is expanded to two-dimensional space to form a two-dimensional matrix, and the two-dimensional matrices of different parameters are superimposed together to form an equipment parameter map.

[0023] Preferably, the equipment parameter diagram is input into a trained equipment health level assessment model to obtain a preliminary assessment of the equipment health level; the equipment health level includes five levels: normal health, attention health, warning health, fault health, and failure health.

[0024] Preferably, the equipment health level assessment model takes the input equipment parameter map as the starting point, extracts features through multi-layer convolution and pooling operations of the backbone network, and then combines the extracted features in two fully connected layers; subsequently, a feature pyramid network is used in the neck network part to perform multi-scale feature fusion processing on features of different scales, and the detection capability of key area features is improved through the self-attention module; finally, information such as the position of the abnormal parameter box, the confidence score and the health level judgment is output to obtain the equipment health level.

[0025] Preferably, the fault level assessment unit maps the equipment health level to a unified standard fault level according to the equipment health level, and outputs the equipment fault level, including:

[0026] Step 401: collect and process key data of equipment health level to obtain keyword vectors;

[0027] Step 402: The unified fault level system model obtains the fault level of the equipment according to the keyword vector.

[0028] Preferably, the maintenance management unit reproduces the parameters within the time window of the abnormal phenomenon according to the fault level of the equipment, uses an analytical method to compare the parameters of the previous time period, performs attribution analysis, and obtains specific parameters of the abnormality; based on the results of the fault parameter attribution analysis, the relevant equipment maintenance plans and knowledge information are automatically retrieved from the knowledge base to form a formatted equipment maintenance plan.

[0029] A method for equipment health management based on a multi-channel parameter map, applicable to the equipment health management system based on a multi-channel parameter map, comprising the steps of:

[0030] Step 1: The parameter collection unit collects health parameters of the equipment;

[0031] Step 2: The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map;

[0032] Step 3: The health status assessment unit obtains the equipment health level based on the equipment parameter map;

[0033] Step 4: The fault level assessment unit establishes a unified fault level and obtains the equipment fault level according to the equipment health level;

[0034] Step 5: The maintenance management unit performs attribution analysis based on the equipment failure level and obtains an equipment maintenance plan.

[0035] (III) Beneficial Effects: The present invention provides an equipment health management system and method based on a multi-channel parameter graph. First, through a unified data architecture and a cross-industry fault system, it supports rapid access and evaluation management of multiple types of equipment. Second, it generates maintenance recommendations driven by a large language model, achieving an automated closed loop from fault detection to repair solutions, providing a scientific basis and intelligent recommendations for equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a structural schematic diagram of an equipment health management system based on a multi-channel parameter map of the present invention;

[0037] Figure 2 It is a flowchart of the steps of an equipment health management method based on a multi-channel parameter map of the present invention;

[0038] Figure 3 It is a schematic diagram of a flow chart of a specific embodiment of the present invention;

[0039] Figure 4 It is a schematic diagram of the principle of the multi-scale feature fusion module of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with preferred embodiments. More details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can obviously be implemented in a variety of other ways different from the description. Those skilled in the art can make similar generalizations and deductions based on actual application situations without violating the connotation of the present invention. Therefore, the scope of protection of the present invention should not be limited by the content of this specific embodiment.

[0041] The accompanying drawings are schematic diagrams of embodiments of the present invention. It should be noted that the drawings are merely examples and are not drawn to scale, and should not be used to limit the actual scope of protection claimed in the present invention.

[0042] An equipment health management system based on multi-channel parameter graph, such as Figure 1As shown, the system includes a parameter collection unit, a parameter map construction unit, a health status assessment unit, a fault level assessment unit, and a maintenance management unit. The parameter collection unit collects equipment health parameters. The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map. The health status assessment unit determines the equipment health level based on the equipment parameter map. The fault level assessment unit establishes a unified fault level and determines the equipment fault level based on the equipment health level. The maintenance management unit performs attribution analysis based on the equipment fault level to determine the equipment repair and maintenance plan.

[0043] A method for equipment health management based on multi-channel parameter graphs, such as Figure 2 、 Figure 3 As shown, it is applicable to the equipment health management system based on the multi-channel parameter map, which specifically includes the following steps:

[0044] Step 1: The parameter collection unit collects health parameters of the equipment.

[0045] Step 2: The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map.

[0046] Step 3: The health status assessment unit obtains the equipment health level according to the equipment parameter diagram.

[0047] Step 4: The fault level assessment unit establishes a unified fault level and obtains the equipment fault level according to the equipment health level.

[0048] Step 5: The maintenance management unit performs attribution analysis based on the equipment failure level and obtains an equipment maintenance plan.

[0049] More specifically, the equipment health parameters collected by the parameter collection unit in step 1 include dynamic parameters and static parameters. Dynamic parameters are primarily time-series parameters acquired by sensors in real time during equipment operation, such as current, voltage, vibration, operating accuracy, energy consumption, and operating time. Static parameters primarily include parameters that do not change over time, such as the last repair, maintenance, and component replacement time.

[0050] Missing data during data collection can lead to biased reasoning and inaccurate results, as models rely on complete data to accurately identify anomalies. For example, when constructing an equipment parameter map, missing data at certain time points can result in an incomplete image structure, affecting the accuracy of health level assessments. Furthermore, if key parameter data is missing during the fault attribution analysis phase, it can be difficult to accurately locate the root cause of the problem, resulting in inaccurate or incomplete repair recommendations. Therefore, multiple measures must be taken to address the issue of incomplete data. Specific steps include:

[0051] Step 101: Perform integrity check on the collected equipment health parameter data.

[0052] After obtaining the dynamic parameters, the parameter threshold is set to check whether the dynamic parameters are missing or have abnormal values, and the problems found are recorded. Values ​​outside the parameter threshold range will be judged as invalid.

[0053] Specifically, the method for identifying missing or anomaly data involves checking that all expected dynamic parameters are complete at the end of each time window (e.g., 30 seconds). A reasonable range is set for each parameter, and values ​​outside the range are marked as invalid. A collection of labeled data is generated to clearly identify missing or anomaly data points.

[0054] Step 102: interpolate the missing data.

[0055] Various interpolation methods can be used, such as linear interpolation, mean interpolation, and median interpolation. Linear interpolation is simple and suitable for most situations, while mean and median interpolation are suitable for situations with less data fluctuation. Additionally, machine learning models can be used to predict missing values, a method that, while complex, may be more accurate. Furthermore, missing parameters can be derived from other related parameters or supplemented by combining historical device data.

[0056] This method uses linear interpolation, which is suitable for cases where data fluctuations are gentle. It uses the valid data points before and after for interpolation. Mean interpolation and / or median interpolation methods use the mean or median value of other valid data within the same time window to fill in the gaps.

[0057] Step 103: Predict and issue an alarm for abnormal values.

[0058] By analyzing the time series of dynamic parameters, abnormal situations such as sudden changes, trend changes or cyclical fluctuations can be identified, and alarms can be issued in time so that relevant personnel can respond quickly. Common analysis methods for anomaly detection include statistical methods and time series analysis methods. The present invention uses statistical methods (such as standard deviation, Z-score) to identify abnormal points in the data. Time series analysis is used to detect sudden changes, trend changes or cyclical fluctuations, and alarms are issued by setting parameter thresholds. When the parameter exceeds the parameter threshold range, an alarm is triggered, and relevant personnel are notified in real time to check and handle it.

[0059] Step 104: Verify and optimize the completed parameter data.

[0060] After initial interpolation, historical data can be used for verification to ensure that the completed data is consistent with actual conditions. During subsequent data acquisition, continuous optimization of the collection strategy is required, such as increasing the sampling frequency or improving sensor calibration, to reduce the occurrence of incomplete data in the future. Compare the completed dataset with historical records to assess the rationality of the interpolation results. Use cross-validation techniques to ensure the accuracy of model predictions. Analyze the causes of missing data, such as sensor failure or communication delays, and take preventive measures. Adjust the sampling frequency and parameter priority based on the operating characteristics of the equipment to improve parameter data collection efficiency.

[0061] More specifically, the specific operation steps of the parameter map construction unit in step 2 are:

[0062] The main task of the parameter graph construction unit is to convert the time-series equipment health parameters into a time series data structure: equipment parameter graph data according to a certain time length (usually 30 seconds, configurable). Then, the time series data of dynamic parameters is converted into the form of equipment parameter graphs, and the time series data is spatialized, so that the equipment operation status can be analyzed using computer vision technology. The specific operation steps are as follows:

[0063] Step 201: Preprocess the equipment health parameters to obtain normalized parameters.

[0064] The data preprocessing includes data cleaning, filling missing values ​​and standardization. Data cleaning refers to the cleaning and removal of outliers and noise. For example, when the current value at a certain time point significantly exceeds the normal range, it may be necessary to mark it as an abnormality and eliminate it. Filling missing values ​​means that when the data of certain sampling points are missing, they can be filled by interpolation or other methods. Standardization refers to normalizing parameters of different dimensions to the same numerical range (for example, [-1, 1] or [0, 1]).

[0065] Step 202: Divide the normalized parameters according to the time series to obtain equipment parameter graph data.

[0066] The time series of dynamic parameters is divided into windows of specified length. The window size can be configured according to actual needs, such as 30 seconds, 60 seconds, etc., and the sliding step size can be set (for example, sliding every 10 seconds) to generate multiple overlapping windows. For example, with a sampling frequency of 1Hz and a time window length of 30 seconds, each window contains 30 data points. For 60 seconds of data, it can be divided into two non-overlapping windows (0-29 seconds and 30-59 seconds) or multiple overlapping windows (for example, with a sliding step size of 10 seconds).

[0067] The different parameters within each time window are treated as image data for different channels. Since time series are one-dimensional data, and deep learning models (such as YOLO or convolutional neural networks) typically require two-dimensional or three-dimensional data input (for example, width, height, and channels), the image data for different channels must be converted into a two-dimensional matrix. For example, the first channel is current data; the second channel is voltage data; the third channel is temperature data; and the fourth channel is vibration data.

[0068] The number of channels depends on the number of dynamic parameters to be analyzed. For example, consider a 30-second time window containing four dynamic parameters (current, voltage, temperature, and vibration). Each parameter has 30 sampling points within the time window, so each channel is a one-dimensional array of length 30. These four one-dimensional arrays are expanded into a two-dimensional matrix to form an image.

[0069] Split the time series data into several small blocks, each containing K sample points. Assuming the time window T is 30 sample points long, split it into M = 5 blocks, each with K = 6 sample points. Perform a sliding average or other statistical processing (such as maximum or minimum) on the data in each block, mapping the K sample points in each block to a scalar value. This compresses the time series data for each parameter into a new array of length M.

[0070] Suppose parameter 1 is [x1, x2, ..., xt]; parameter 2 is [y1, y2, ..., yt]; parameter 3 is [z1, z2, ..., zt]; parameter 4 is [w1, w2, ..., wt]. For each parameter, t is set to 30 sampling points, and the data is divided into M = 5 blocks, so the length of each block is K = 6. In parameter 1, Block1 = [x1, x2, ..., x6], Block2 = [x7, x8, ..., x12], and so on. For each block, calculate the sliding average of 6 sampling points. The average value of the i-th block in parameter 1 is:

[0071]

[0072] Among them, a i represents the average value of the i-th block; K represents the number of sampling points; j represents the current single sampling point.

[0073] Step 203: normalize the equipment parameter map data to obtain normalized image data.

[0074] To meet the input requirements of computer vision models, image data from all channels must be normalized to produce normalized image data. For example, if grayscale values ​​from 0 to 255 are used to represent normalized image data, the normalized parameter values ​​for each equipment health parameter must be mapped to this normalized image data range. Alternatively, if a standardization method is used (e.g., a mean of 0 and a standard deviation of 1), the equipment parameter map data must be normalized.

[0075] Step 204: Combine the normalized image data into a two-dimensional equipment parameter map.

[0076] Specific implementations can be: arranging the normalized image data within each time window into a matrix (rows represent time, columns represent parameters). Alternatively, the normalized image data can be expanded into a higher-dimensional space (e.g., generating two-dimensional spatial information through sliding average or other methods). For example, the time window length is 30 seconds and contains four dynamic parameters. Arrange the 30 sample points of each parameter into a one-dimensional array, and then combine them into a four-channel image (shape 30 × 1 × 4).

[0077] The normalized image data is then expanded to a two-dimensional space, where a two-dimensional matrix can be generated using sliding averaging or other methods. This involves arranging the new arrays of each parameter into an M×M two-dimensional matrix (e.g., by repetition or expansion). This ultimately results in a two-dimensional matrix with a shape of (M,M). These four two-dimensional matrixes are then superimposed to form the equipment parameter map. For example, if each parameter corresponds to a channel, the resulting three-dimensional array has a shape of (M,M,4) (width, height, number of channels).

[0078] Arrange the processed array of each parameter (length M=5) into a two-dimensional matrix. This can be done by repeating or expanding. For example, the two-dimensional matrix of parameter 1 is (The specific method depends on the implementation details). In practice, how to arrange these five averages depends on the specific scenario requirements. For example: if you need to maintain a simple structure, you can repeat the last value. If you need higher resolution, you can interpolate the data. If you need a larger matrix size, you can fill it by expanding or sliding the window. If you want to keep the original five averages, you can arrange them into a 2×3 matrix (repeating the last value):

[0079]

[0080] The two-dimensional matrices of these four parameters are superimposed together to form an equipment parameter map: Image = [A, B, C, D], where A, B, C, and D are four two-dimensional matrices respectively.

[0081] The final generated equipment parameter map can be used as input to deep learning models (such as YOLO, convolutional neural networks, etc.) for subsequent feature extraction and classification tasks.

[0082] More specifically, the specific operation steps of the health status assessment unit in step 3 are as follows:

[0083] The equipment parameter graph is input into the trained equipment health level assessment model to obtain a preliminary assessment of the equipment health level. The equipment health level includes five levels: normal health, attention health, warning health, fault health, and failure health.

[0084] The equipment health level assessment model, based on the YOLO image recognition and detection algorithm, consists of a backbone and a core network. The backbone consists of four convolutional layers, three pooling layers, and two fully connected layers. The core network uses an FPN feature pyramid network structure and incorporates a self-attention mechanism. The model also incorporates features related to impact consequences (such as equipment type and operating environment), thereby enhancing the semantic richness of the equipment context information in the fault level assessment unit.

[0085] Specifically, the equipment health level assessment model uses an input equipment parameter map as its starting point. The backbone network extracts features through multiple layers of convolution and pooling. Two fully connected layers then combine these extracted features. The neck network then uses a feature pyramid network (FPN) to fuse features at different scales, and a self-attention module enhances the detection of key regional features. Ultimately, the model outputs information such as the location of the abnormal parameter box, a confidence score, and a health level judgment, completing the equipment health level assessment.

[0086] In the self-attention module, the self-attention weight calculation formula of the feature map F is:

[0087]

[0088] Among them, Q is the query matrix; K is the key matrix; V is the value matrix; d k is the scaling factor, equal to the dimension of Q.

[0089] The self-attention module can enhance the attention of the equipment health level assessment model to different feature areas, and enhance the ability to capture key features at different feature scales, thereby more accurately identifying abnormal information in the equipment parameter map and improving the accuracy of feature selection.

[0090] Since the equipment health level assessment model is not accurate enough in detecting low-density and small-area phenomena, the convolutional structure of the basic model is difficult to effectively extract the spatial features of small-area phenomena. Therefore, the equipment health level assessment model is optimized based on the residual network, and a multi-scale feature fusion module is introduced to improve the accuracy of the equipment health level assessment model in detecting small-scale phenomena. The optimization of the equipment health level assessment model by the multi-scale feature fusion module is as follows: Figure 4 shown.

[0091] The multi-scale feature fusion module is constructed from different feature layers (F i , F j , F k ) to extract information to ensure that features of multiple scales can participate in the detection. In the weighted fusion calculation stage, the multi-scale feature fusion module performs weighted merging of feature layers according to weight coefficients (α, β, γ) to obtain fusion features (F out ), as shown below:

[0092] F out =α·F i +β·F j +γ·F k ,

[0093] Among them, α, β, and γ are learnable weighting coefficients that ensure the effective fusion of features at different scales.

[0094] The multi-scale feature fusion module improves the detection accuracy of the equipment health level assessment model in small areas and sparse scenes, and can better identify abnormal parameter map areas.

[0095] The loss function of the equipment health level assessment model is calculated based on the output information of the equipment health level assessment model, and the equipment health level assessment model is optimized so that the equipment health level assessment model can efficiently complete the preliminary judgment of the equipment health level.

[0096] The loss function of the equipment health level assessment model includes regression loss, classification loss, and confidence loss. The total loss L is expressed as follows:

[0097]

[0098]

[0099] Among them, λ coord ,λ conf ,λ class They are regression loss weight, confidence loss weight, and classification loss weight respectively; x i ,y iare the predicted coordinates of the anomaly region box; are the real coordinates of the abnormal phenomenon area frame; p j is the confidence prediction score; is the true confidence score; c k is the predicted probability distribution of equipment health level; is the true probability distribution of the equipment health level; i is the current predicted abnormal phenomenon area box; N is the number of area boxes that need to predict abnormal phenomena.

[0100] More specifically, the specific operation steps of the fault level assessment unit in step 4 are:

[0101] The fault level assessment unit maps the equipment health level to a unified standard fault level according to the equipment health level, and outputs the equipment fault level.

[0102] Specifically, the equipment health level assessment model first outputs the abnormal parameter box location, confidence score, and equipment health level. The preliminary health level and equipment context information are then input into a unified fault level system model, ultimately outputting a standardized fault level for user or management system decision-making.

[0103] The impact of a failure varies by industry and device type, and is crucial information that cannot be ignored for every failure. Failure levels are expressed differently across industries and device types. For example, some equipment failure levels are categorized from high to low, ranging from "severe, moderate, minor, or no impact," while some industries are divided into "fatal, emergency, warning, and attention levels." Corresponding the failure levels of different devices to create a unified failure level system is key. Therefore, starting from the impact dimension, the unified failure level system model designs a system centered around safety, operational efficiency, economy, and environmental factors. Each dimension is further divided into cross-impact levels, and ultimately, the overall standardized failure level is determined based on these comprehensive indicators.

[0104] The impact level is designed into four main levels: fatal, emergency, warning and concern.

[0105] In addition to the impact level, specific consequences should also be provided. This is because different industries and equipment types define impact levels differently. For example, "serious" may mean equipment downtime in one industry, while in another it may simply mean reduced efficiency. Therefore, adding detailed descriptions of the impact and consequences helps quickly identify the impact and consequences of a failure, facilitating quick decision-making by managers.

[0106] This invention uses deep learning to solve the problem of inconsistent fault level standards for different equipment. The specific implementation steps are as follows:

[0107] Step 401: collect and process key data of equipment health level to obtain keyword vectors.

[0108] The key data include operation and maintenance records of different equipment types and different fault level standards, as well as equipment manuals.

[0109] Clean and preprocess key data, including sentence segmentation, keyword extraction, word vectorization, etc.

[0110] The sentence segmentation is to segment the output equipment health level and impact level based on semantics. When selecting a deep learning model for word segmentation, it is necessary to consider the relationship between paragraphs and sentences in the input content. Because Transformer can more effectively handle long-distance dependencies and has stronger semantic understanding capabilities, the present invention prefers Transformer as the word segmentation model.

[0111] The segmented attention mechanism segments the fault description into independent sentences or paragraphs and calculates an attention weight for each paragraph, helping to identify specific paragraphs describing the fault's impact and consequences. A cross-attention mechanism is introduced between two sentences (for example, a sentence describing the impact and a sentence describing the consequences) to learn the relationship between them. Fault descriptions can be divided into multiple levels, such as symptoms, causes, impacts, and consequences. A multi-layer self-attention mechanism is used to learn the relationships between these levels, guiding the model to focus on information related to impacts and consequences.

[0112] Keyword extraction involves extracting keywords after word segmentation based on sentence meaning. For example, keywords related to "impact" and "consequences" such as "production suspension," "reduced efficiency," "data loss," and "safety hazard" are extracted from the fault description and used as keywords for keyword extraction.

[0113] Word vectorization involves vectorizing keywords to generate corresponding keyword vectors. Keyword vectors are constructed using methods such as Word2Vec or GloVe, focusing on the semantic similarity between words related to fault impact and consequences. These associated word vectors can be used as weights in the attention mechanism to guide the model's focus on related word combinations.

[0114] More specifically, the present invention constructs a keyword training dataset based on keyword training data, which uses fault description text as input and outputs corresponding fault level labels. This keyword training dataset is used to train the unified fault level system model. A mapping relationship exists between the keyword training dataset and the fault level, such that the presence of a particular keyword corresponds to a particular fault level. This can be achieved by adjusting model parameters to ensure that the predicted results are as close as possible to the true labels, and / or by employing cross-validation or other methods to select the optimal model parameters. The specific method is not limited.

[0115] Step 402: The unified fault level system model obtains the fault level of the equipment according to the keyword vector.

[0116] The unified fault level system model predicts the fault level label corresponding to the fault description based on the mapping relationship corresponding to the keyword vector.

[0117] More specifically, the specific operations of the maintenance management unit in step 5 are:

[0118] Based on the equipment's fault level, the parameters within the time window of the abnormal phenomenon are reproduced and compared with the parameters of the previous time period using analytical methods to conduct attribution analysis. The specific parameters of the abnormality can be obtained, such as current, voltage, vibration, operating accuracy, energy consumption, and other single or multiple indicators. This attribution analysis provides a valuable basis for subsequent fault repair and equipment health management.

[0119] For example: Using the analytical method to compare with the previous period, it is found that the current has increased by x% and the vibration amplitude has increased by y mm / s, and the change trends of the two show a strong positive correlation (the correlation can be calculated using methods such as the Pearson correlation coefficient). This indicates that there may be a local short circuit in the motor winding, which in turn causes the current to increase, and at the same time causes the motor to run unbalanced and produce abnormal vibration.

[0120] Based on fastGPT, ollama, and deepseek-r1:14b, we extensively collected data on equipment maintenance history, equipment technical manuals, procedures and specifications, and historical maintenance cases to build an equipment maintenance knowledge base based on a large language model. This knowledge base includes structural diagrams, operating manuals, common fault diagnosis and repair solutions, and maintenance plans for various equipment.

[0121] Based on the results of the fault parameter attribution analysis, a standardized prompt is automatically generated, and the FastGPT access API interface is called to obtain the relevant equipment maintenance plans and knowledge information retrieved by the large model from the knowledge base, forming a formatted equipment maintenance plan and outputting it to the maintenance personnel.

[0122] The equipment maintenance plan includes maintenance time, maintenance type recommendations, maintenance technical requirements, etc.

[0123] This invention proposes an equipment health management system and method based on a multi-channel parameter graph. First, through a unified data architecture and cross-industry fault system, it supports rapid access and evaluation management for multiple types of equipment. Second, it generates maintenance recommendations driven by a large language model, achieving an automated closed-loop from fault detection to repair plan, providing a scientific basis and intelligent recommendations for equipment repair and maintenance.

[0124] The above content is an explanation of the preferred embodiments of the present invention, which can help those skilled in the art to more fully understand the technical solutions of the present invention. However, these embodiments are merely illustrative, and it cannot be determined that the specific implementation methods of the present invention are limited to the description of these embodiments. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions and transformations can be made, which should be deemed to fall within the scope of protection of the present invention.

Claims

1. An equipment health management system based on a multi-channel parameter map, characterized in that: It includes a parameter collection unit, a parameter map construction unit, a health status assessment unit, a fault level assessment unit, and a maintenance management unit; the parameter collection unit collects health parameters of the equipment; The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map; The health status assessment unit obtains the equipment health level based on the equipment parameter map; The fault level assessment unit establishes a unified fault level and obtains the equipment fault level based on the equipment health level; The maintenance management unit conducts attribution analysis based on the equipment failure level and obtains the equipment maintenance plan.

2. The equipment health management system based on a multi-channel parameter map according to claim 1, characterized in that: The equipment health parameters include dynamic parameters and static parameters. The dynamic parameters are mainly time-series parameters acquired by sensors in real time during the operation of the equipment. Missing data during data collection can lead to inference bias. Measures must be taken to address the problem of incomplete data, including: Step 101: Perform integrity check on the collected equipment health parameter data; Step 102: interpolate the missing data. Step 103: predict and issue an alarm for abnormal values; Step 104: Verify and optimize the completed parameter data.

3. The equipment health management system based on a multi-channel parameter map according to claim 1, characterized in that: The parameter graph construction unit converts the time-series equipment health parameters into time series data according to a certain time length: equipment parameter graph data, and then the time series data is converted into the equipment parameter graph form to spatialize the time series data.

4. The equipment health management system based on a multi-channel parameter map according to claim 3, characterized in that: The parameter map construction unit specifically includes: step 201, performing data preprocessing on equipment health parameters to obtain normalized parameters; Step 202: Divide the normalized parameters according to the time series to obtain equipment parameter graph data; Step 203: normalize the equipment parameter map data to obtain normalized image data; Step 204: Combine the normalized image data into a two-dimensional equipment parameter map.

5. The equipment health management system based on a multi-channel parameter map according to claim 4, characterized in that: The time series of dynamic parameters is divided according to windows of specified length. Different parameters in each time window are used as image data of different channels. The data of all channels are normalized to obtain normalized image data. The normalized image data is expanded to two-dimensional space to form a two-dimensional matrix. The two-dimensional matrices of different parameters are superimposed together to form an equipment parameter map.

6. The equipment health management system based on a multi-channel parameter map according to claim 1, characterized in that: The equipment parameter diagram is input into the trained equipment health level assessment model to obtain a preliminary assessment of the equipment health level; the equipment health level includes five levels: normal health, attention health, warning health, fault health, and failure health.

7. The equipment health management system based on multi-channel parameter graph according to claim 6, characterized in that: The equipment health level assessment model uses the input equipment parameter map as the starting point, extracts features through multi-layer convolution and pooling operations in the backbone network, and then combines the extracted features in two fully connected layers. Subsequently, the feature pyramid network is used in the neck network part to perform multi-scale feature fusion processing on features of different scales, and the self-attention module is used to improve the detection ability of key area features; finally, the abnormal parameter box position, confidence score, health level judgment and other information are output to obtain the equipment health level.

8. The equipment health management system based on a multi-channel parameter map according to claim 1, characterized in that: The fault level assessment unit maps the equipment health level to a unified standard fault level according to the equipment health level and outputs the equipment fault level, including: Step 401: collect and process key data of equipment health level to obtain keyword vectors; Step 402: The unified fault level system model obtains the fault level of the equipment according to the keyword vector.

9. The equipment health management system based on multi-channel parameter graph according to claim 1, characterized in that: The maintenance management unit reproduces the parameters within the abnormal phenomenon time window according to the fault level of the equipment, uses analytical methods to compare the parameters of the previous time period, performs attribution analysis, and obtains the specific parameters of the abnormality; Based on the results of fault parameter attribution analysis, relevant equipment maintenance plans and knowledge information are automatically retrieved from the knowledge base to form a formatted equipment maintenance plan.

10. A method for equipment health management based on a multi-channel parameter map, applicable to the equipment health management system based on a multi-channel parameter map, characterized in that: Including steps: Step 1: The parameter collection unit collects health parameters of the equipment; Step 2: The parameter map construction unit converts the collected equipment health parameters into equipment parameter map data and constructs the equipment parameter map; Step 3: The health status assessment unit obtains the equipment health level based on the equipment parameter map; Step 4: The fault level assessment unit establishes a unified fault level and obtains the equipment fault level according to the equipment health level; Step 5: The maintenance management unit performs attribution analysis based on the equipment failure level and obtains an equipment maintenance plan.

Images