Lightweight equipment fault prediction system based on multi-dimensional data
Through a lightweight equipment fault prediction system with multi-dimensional data, semi-supervised learning and deep learning technology can capture the device status in real time, solving the problems of data annotation and fault diagnosis accuracy of traditional systems, improving the warning response speed and maintenance decision-making Scientific, reducing maintenance costs.
Patent Information
- Application Number
- CN202410931901.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional lightweight equipment fault prediction systems have data labeling inconsistencies and errors when processing unstructured or semi-structured data, and fail to make full use of the dependencies between device components, resulting in the accuracy of fault diagnosis results and early warning delays, which increases the Maintenance cost.
Adopt a lightweight equipment fault prediction system based on multi-dimensional data, through data acquisition, label processing, fault association identification, abnormal trend analysis and fault prediction early warning modules, using semi-supervised learning, graph neural network, Transformer and long-term short-term memory network and other technologies. Capture device status in real time, extract abnormal modes, and optimize early warning and maintenance decisions.
It improves the automation level and efficiency of data processing, enhances the accuracy of fault diagnosis, improves the response speed and accuracy of early warnings, and optimizes maintenance plans and costs.
Smart Images

Figure CN118940167A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field, and in particular to a lightweight equipment fault prediction system based on multi-dimensional data. Background Art
[0002] The field of fault prediction technology involves predicting possible fault conditions in equipment or systems through data analysis and algorithmic models. Fault prediction technology identifies potential signs of failure by monitoring data during equipment operation, such as sensor data, operating parameters, and performance indicators. Specific methods include machine learning algorithms, statistical analysis, and model training to achieve real-time monitoring and prediction of equipment health status. Fault prediction technology can not only detect problems that may cause equipment downtime or performance degradation in advance, but also optimize maintenance plans and resource utilization, reduce maintenance costs, and improve equipment reliability.
[0003] Among them, the lightweight equipment fault prediction system aims to develop an efficient and low-cost fault prediction solution for lightweight equipment such as portable electronic devices, mobile communication equipment or small industrial equipment. The system combines data analysis technology and real-time monitoring methods to predict possible failures through real-time analysis of equipment operation data, and provide timely warning information and maintenance suggestions. Its application areas cover lightweight equipment that requires high reliability and long-term operation, aiming to optimize the operating efficiency of the equipment and reduce maintenance costs.
[0004] Traditional prediction systems rely heavily on data monitoring when dealing with lightweight equipment fault predictions, and show certain limitations when dealing with unstructured or semi-structured data. Traditional systems are insufficient in automating and intelligently processing unlabeled data, resulting in inconsistencies and errors in data labeling, which affects the final fault diagnosis results. Traditional systems fail to fully utilize the dependencies between equipment components, ignore the propagation paths of fault modes between different components, limit the depth and breadth of fault root cause analysis, and have defects in dynamically capturing equipment status changes and performing real-time fault predictions, resulting in delays and accuracy issues in fault warnings, increasing maintenance costs and affecting the overall operating efficiency of the equipment. Summary of the invention
[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a lightweight equipment fault prediction system based on multi-dimensional data.
[0006] In order to achieve the above object, the present invention adopts the following technical solution: a lightweight equipment fault prediction system based on multidimensional data, the system comprising:
[0007] The data acquisition module captures the operating status data of the lightweight device based on the operating environment of the lightweight device, and performs timestamp marking and format standardization on the data to obtain a basic status data set;
[0008] The data label processing module processes the unlabeled data based on the basic state data set by using a semi-supervised learning method and performs a consistency check on the labels to obtain verified data labels;
[0009] The fault association identification module uses a graph neural network to model the component functions and failure modes of the lightweight equipment based on the verified data labels, captures the propagation path of the fault, and obtains fault association information;
[0010] The abnormal trend analysis module analyzes the operation data of the lightweight equipment using Transformer and long short-term memory network based on the fault association information, verified data labels and basic status data set, extracts the trend changes and abnormal patterns of the data, and obtains the abnormal trend analysis results;
[0011] The fault prediction and warning module predicts the time and type of lightweight equipment failure based on the abnormal trend analysis results, adjusts the alarm parameters, optimizes the accuracy and response speed of the alarm, and obtains abnormal warning information;
[0012] The maintenance optimization decision module evaluates the current maintenance requirements of the equipment, evaluates the priority of maintenance activities, adjusts the maintenance sequence and maintenance time, and obtains maintenance decision information based on the abnormal warning information.
[0013] As a further solution of the present invention, the basic status data set includes time synchronization data, format unified data and noise filtered data, the verified data labels include normal labels, abnormal labels and misjudgment labels, the fault association information includes graphical equipment structure, equipment component dependencies and predicted functional failure points, the abnormal trend analysis results include time series trend data, abnormal mode indicators and potential fault signals, the abnormal warning information includes status monitoring data, fault mode identification and warning signal information, and the maintenance decision information includes cost-effectiveness information, maintenance strategy information and maintenance time points.
[0014] As a further solution of the present invention, the data acquisition module includes:
[0015] The time stamp submodule captures the vibration, temperature and current data of the lightweight equipment based on its operating environment, performs time stamping, and generates time series data for each data record corresponding to the date and time.
[0016] The format standardization submodule performs data format unification processing based on the time series data, converts the multi-source data into a unified data format, and obtains unified format data;
[0017] The noise filtering submodule performs noise filtering on the data stream based on the unified format data, removes electrical interference and atypical data noise points through filtering, optimizes the accuracy and reliability of the data, and obtains a basic status data set.
[0018] As a further solution of the present invention, the data label processing module includes:
[0019] The data classification submodule performs machine learning classification on the unlabeled data based on the basic state data set, identifies patterns and categories in the data, and obtains category identification data;
[0020] The labeling processing submodule uses a semi-supervised learning method to label the data based on the category recognition data, assigns corresponding labels to each data point, including normal, abnormal and warning, and converts them into a machine-readable form through encoding technology to optimize the consistency and operability of the data set to obtain processed label data;
[0021] The consistency check submodule performs consistency check on the label based on the processed label data, adopts statistical methods and logic verification to test the consistency and accuracy of the data label, verifies that the label reflects the state and characteristics of the data, and obtains the verified data label.
[0022] As a further solution of the present invention, the fault association identification module includes:
[0023] The relationship analysis submodule analyzes the functional associations between various components of the lightweight equipment based on the verified data tags, identifies the interactions and dependencies between component functions through statistical correlation analysis of the data, and obtains component function mapping information;
[0024] The performance index extraction submodule identifies and extracts key performance indicators based on the component function mapping information, including abnormal temperature and current mutation associated with the fault, and identifies the performance indicators associated with the fault mode by evaluating the performance indicators, thereby obtaining key performance indicator information;
[0025] Based on the key performance indicator information, the dependency building submodule constructs a dependency model between lightweight device components through graph neural network technology, and obtains fault correlation information by adjusting the weights of nodes and edges to identify the impact and fault propagation path between components.
[0026] As a further solution of the present invention, the abnormal trend analysis module includes:
[0027] The sequence analysis submodule performs time series analysis on the operation data of the equipment through a long short-term memory network based on the fault association information, the verified data labels and the basic state data set, identifies the time dependency and periodic changes of the data, extracts key time series features, and obtains a time dependency feature set;
[0028] The pattern recognition submodule performs pattern recognition on the data based on the time-dependent feature set through a support vector machine, analyzes and classifies the normal operation mode and multiple abnormal modes of the equipment, and obtains the equipment mode classification result;
[0029] Based on the equipment mode classification results, the trend identification submodule analyzes long-term data trends and abnormal behavior development through Transformer technology, predicts the behavior of the equipment in the future, extracts the change trend and abnormal pattern of the equipment status, and obtains the abnormal trend analysis results.
[0030] As a further solution of the present invention, the support vector machine is according to the formula:
[0031]
[0032] Calculate the classification results and obtain the device mode classification results, where α i is the Lagrange multiplier, y i is the class label of the i-th data point, x i is the i-th support vector, x is the input feature, γ is the width parameter of the function, e is the logarithm of the natural logarithm, λ is the weight coefficient of the hyperbolic tangent function, β is the scale parameter of the hyperbolic tangent function, b is the bias term, n represents the total number of support vectors, and f(x) is the classification function.
[0033] As a further solution of the present invention, the fault prediction and warning module includes:
[0034] The state monitoring submodule monitors the operating state of the lightweight equipment in real time based on the abnormal trend analysis results, evaluates the deviation of the data from the normal operating mode, identifies performance degradation and abnormal conditions, and obtains real-time state information;
[0035] The fault analysis submodule analyzes potential faults based on the real-time status information, predicts the type and occurrence time of the fault through a machine learning model, and obtains fault analysis information;
[0036] The warning generation submodule adjusts warning parameters based on the fault analysis information, the warning parameters including the threshold of fault severity and the response time, optimizes the accuracy and response speed of the alarm, and generates abnormal warning information.
[0037] As a further solution of the present invention, the maintenance optimization decision module includes:
[0038] The maintenance demand analysis submodule analyzes the maintenance demand of lightweight equipment based on the abnormal warning information, combined with the historical maintenance records and fault data of lightweight equipment, identifies key maintenance points and the urgency of maintenance, formulates a maintenance task list, and obtains the maintenance demand analysis results;
[0039] The maintenance priority assessment submodule uses a multi-criteria decision analysis method based on the maintenance demand analysis results, combines the potential impact of the failure, equipment criticality and operation interruption cost, evaluates the priority of multiple maintenance activities, optimizes the continuous operation of key equipment, and obtains maintenance priority information;
[0040] The maintenance strategy adjustment submodule adjusts the time and resource allocation of the maintenance plan based on the maintenance priority information, according to the current operating status and maintenance cost of the lightweight equipment, optimizes the cost efficiency of maintenance, and obtains maintenance decision information.
[0041] As a further solution of the present invention, the multi-criteria decision analysis method is according to the formula:
[0042]
[0043] Calculate the priority score of the maintenance activity and obtain the maintenance priority information, where:
[0044] S i is the total score of the ith maintenance activity, w j is the weight of the jth criterion, c ij is the score of the ith maintenance activity under the jth criterion, m is the total number of criteria, k is the slope parameter of the curve, τ j is the score threshold under the jth criterion, and e is the base of the natural logarithm.
[0045] Compared with the prior art, the advantages and positive effects of the present invention are:
[0046] In the present invention, by real-time capture and standardization of data, the availability and consistency of data are enhanced, a semi-supervised learning method is used to process unlabeled data, the automation level and efficiency of data processing are improved, data labeling is made more accurate, and the accuracy of fault diagnosis is enhanced. By capturing the dependencies between components and the fault propagation path, an in-depth understanding of the root cause and propagation mode of the fault is obtained, and Transformer and long short-term memory networks are used to extract data trends and abnormal patterns, so that fault prediction can dynamically reflect the operating status of the equipment, improve the response speed of the early warning and the accuracy of the alarm, make maintenance decisions more scientific, and effectively optimize maintenance plans and costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a system flow chart of the present invention;
[0048] Figure 2 It is a schematic diagram of the system framework of the present invention;
[0049] Figure 3 It is a flow chart of the data acquisition module of the present invention;
[0050] Figure 4 This is a flow chart of the data tag processing module of the present invention;
[0051] Figure 5 A flowchart of a fault association identification module of the present invention;
[0052] Figure 6 It is a flow chart of the abnormal trend analysis module of the present invention;
[0053] Figure 7 This is a flow chart of the fault prediction and early warning module of the present invention;
[0054] Figure 8 A flowchart of the maintenance optimization decision module of the present invention. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0056] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.
[0057] Embodiment 1
[0058] See also Figure 1 , a lightweight equipment fault prediction system based on multi-dimensional data, the system includes:
[0059] The data acquisition module captures the operating status data of lightweight equipment based on the operating environment of lightweight equipment. The operating status data includes the vibration, temperature and current of the equipment, and performs timestamp marking and format standardization on the data to unify the data format, and performs noise filtering on the data stream to optimize the data quality and obtain the basic status data set;
[0060] The data label processing module processes unlabeled data based on the basic state data set by using a semi-supervised learning method, classifies and labels the data, and performs consistency checks on the labels to verify the correctness of the data labels and obtain verified data labels.
[0061] The fault correlation identification module uses graph neural networks to model the component functions and failure modes of lightweight equipment based on verified data labels. By adjusting the weights of nodes and edges, it simulates the dependencies between components, captures the propagation path of faults, and obtains fault correlation information.
[0062] The abnormal trend analysis module uses Transformer and long short-term memory network to analyze the operation data of lightweight equipment based on fault correlation information, verified data labels and basic status data sets, extracts trend changes and abnormal patterns of data, and obtains abnormal trend analysis results;
[0063] The fault prediction and warning module predicts the time and type of lightweight equipment failure based on the abnormal trend analysis results, analyzes potential failures, adjusts alarm parameters, optimizes the accuracy and response speed of the alarm, and obtains abnormal warning information;
[0064] The maintenance optimization decision module evaluates the current maintenance needs of the equipment based on abnormal warning information, combined with the historical maintenance records and fault data of lightweight equipment, and evaluates the priority of maintenance activities according to the potential impact of the failure. It also adjusts the maintenance sequence and maintenance time according to the maintenance cost to obtain maintenance decision information;
[0065] The basic status data set includes time-synchronized data, format-unified data, and noise-filtered data. The verified data labels include normal labels, abnormal labels, and misjudgment labels. The fault-related information includes graphical equipment structure, equipment component dependencies, and predicted functional failure points. The abnormal trend analysis results include time series trend data, abnormal pattern indicators, and potential failure signals. The abnormal warning information includes status monitoring data, failure pattern recognition, and warning signal information. The maintenance decision information includes cost-effectiveness information, maintenance strategy information, and maintenance time points.
[0066] See also Figure 2 and Figure 3 ,The data acquisition module includes a time marking submodule, a format standardization submodule, and a noise filtering submodule;
[0067] The time stamp submodule captures the vibration, temperature and current data of the lightweight equipment based on the operating environment of the lightweight equipment, performs time stamping, and records the corresponding date and time for each data. The specific process of generating time series data is as follows;
[0068] The time stamp submodule captures the vibration, temperature and current data of lightweight equipment operation based on the operating environment of lightweight equipment, and converts analog signals into digital signals. After conversion, the current date and time are added to each data record to form a timestamp. The time difference obtained by subtracting the data entry time from the system base time is used as the timestamp. The formula is expressed as: "Timestamp = System time - Data reception time", generating time series data.
[0069] The format standardization submodule performs data format unification based on time series data, converts multi-source data into a unified data format, and obtains the unified format data in the following specific processes:
[0070] The format standardization submodule identifies and converts the data formats of different data sources into a preset unified format based on time series data. During the processing, the original format of each data is converted to a unified template through a mapping table, including timestamp, data type and measurement value. For the processing of missing fields, default values are used to fill in the filling method: "Default value = measurement average - standard deviation" to verify the integrity and accuracy of the converted data and obtain unified format data.
[0071] The noise filtering submodule performs noise filtering on the data stream based on the unified format data, removes electrical interference and atypical data noise, optimizes the accuracy and reliability of the data, and obtains the basic status data set process as follows;
[0072] The noise filtering submodule defines the threshold parameters of noise and interference based on the unified format data and filters the data. The noise threshold of the current data is set as: "Threshold = average current ± 10%". If the current value of each data exceeds this range, it is marked as noise. For vibration and temperature data, moving average filtering is applied, and the calculation method is: "Filter result = (previous data + current data + next data) / 3". The abnormal detection of atypical data points is identified by calculating the deviation of the data point from the moving average, and the formula is: "Deviation = data point - filter result". Through calculation, the electrical interference and atypical noise in the data are removed to obtain the basic state data set.
[0073] See also Figure 2 and Figure 4 ,The data label processing module includes a data classification submodule, a label processing submodule, and a consistency checking submodule;
[0074] The data classification submodule performs machine learning classification on unlabeled data based on the basic state data set to identify patterns and categories in the data. The specific process of obtaining category recognition data is as follows:
[0075] The data classification submodule classifies the unlabeled data based on the basic state data set, defines feature vectors, including vibration, temperature, current and other parameters, uses the classification algorithm to identify patterns and categories in the data, and standardizes each feature. The formula is expressed as: "Standardized value = (original value - mean) / standard deviation". The standardized data is trained using the classification algorithm. The classification result is determined by calculating the distance between each data point and the center of each category. The formula is: "Distance = √((vibration difference)^2+(temperature difference)^2+(current difference)^2)". According to the minimum distance principle, a category label is assigned to each data point to obtain category recognition data.
[0076] The labeling processing submodule uses a semi-supervised learning method to label data based on category recognition data, assigning corresponding labels to each data point, including normal, abnormal, and warning. Through encoding technology, it is converted into a machine-readable form to optimize the consistency and operability of the data set. The specific process of obtaining the processed label data is as follows;
[0077] The labeling processing submodule uses a semi-supervised learning method to label the data based on category recognition data. Some data points are selected for manual labeling, and the model is trained through data points with known labels. The trained model is used to predict unlabeled data points. The specific calculation formula is: "Prediction label = (model output)". According to the confidence of the model output, the label assignment rule is defined. Confidence levels above 0.8 are assigned as normal, those below 0.5 are assigned as abnormal, and those between 0.5 and 0.8 are assigned as warnings. The formula is expressed as: "Label = Normal (confidence > 0.8); Abnormal (confidence < 0.5); Warning (0.5 ≤ confidence ≤ 0.8)". Through encoding technology, the label is converted into a form that can be interpreted by the machine, for example: normal is 0, abnormal is 1,
[0078] The warning is 2, optimizing the consistency and operability of the dataset and obtaining processed labeled data.
[0079] The consistency check submodule performs consistency check on the label based on the processed label data, uses statistical methods and logic verification to test the consistency and accuracy of the data label, and verifies that the label reflects the state and characteristics of the data. The specific process of obtaining the verified data label is as follows;
[0080] The consistency check submodule performs consistency checks on the labels based on the processed label data. It uses statistical methods to analyze the distribution of each type of label and calculates the frequency of each type of label. The formula is: "Frequency = number of labels / total number". The logic verification method is used to compare the label and its features of each data point to verify whether the label reflects the true state of the data. The correlation between each label category and the data feature is calculated. The formula is: "Correlation = Σ (label value * feature value) / total number". If the correlation is high, it means that the label allocation is correct. Otherwise, the label needs to be readjusted to obtain the verified data label.
[0081] See also Figure 2 and Figure 5 ,The fault association identification module includes a relationship parsing submodule, a performance indicator extraction submodule, and a dependency relationship building submodule;
[0082] The relationship analysis submodule analyzes the functional associations between various components of lightweight equipment based on the verified data labels, identifies the interactions and dependencies between component functions through statistical correlation analysis of the data, and obtains the component function mapping information in the following specific process:
[0083] The relationship analysis submodule analyzes the functional associations between various components of lightweight equipment based on the verified data labels, performs statistical analysis on the data of each component, and calculates the correlation coefficients between the components. The formula is: "Correlation coefficient = ∑((component 1 data - component 1 mean) * (component 2 data - component 2 mean)) / (number of data points - 1)". According to the calculation results, the correlation matrix is drawn to identify the interactions and dependencies between components. The statistical results show that there are stronger functional associations between components with higher correlation coefficients, and the component function mapping information is obtained.
[0084] The performance index extraction submodule identifies and extracts key performance indicators based on component function mapping information, including abnormal temperature and current mutations associated with faults. By evaluating the performance indicators, the performance indicators associated with the fault mode are identified. The specific process of obtaining key performance indicator information is as follows;
[0085] The performance index extraction submodule identifies and extracts key performance indicators based on component function mapping information, and screens out abnormal temperature and current mutation points associated with faults. The formula is: "abnormal point = data point value - average value > threshold value", and uses the evaluation model to analyze these abnormal points. The formula is: "evaluation value = abnormal temperature + current mutation". By sorting and analyzing the evaluation values, the performance indicators associated with the fault mode are identified, and the key performance indicator information is obtained.
[0086] The dependency building submodule builds a dependency model between lightweight equipment components based on key performance indicator information through graph neural network technology. By adjusting the weights of nodes and edges, the impact between components and fault propagation paths are identified. The specific process of obtaining fault association information is as follows;
[0087] The dependency building submodule builds a dependency model between lightweight device components based on key performance indicator information, defines device components as nodes, performance indicators as edge weights, and sets the initial weights to the numerical values of performance indicators. By gradually adjusting the weights of nodes and edges, the calculation formula is expressed as: "Weight adjustment = current weight * adjustment coefficient + new data contribution value". According to the adjusted weights, the impact between components and the fault propagation path are identified, and finally the fault association information is obtained.
[0088] See also Figure 2 and Figure 6 ,The abnormal trend analysis module includes a sequence analysis submodule, a pattern recognition submodule, and a trend recognition submodule;
[0089] The sequence analysis submodule performs time series analysis on the equipment operation data through the long short-term memory network based on the fault association information, the verified data labels and the basic status data set, identifies the time dependency and periodic changes of the data, extracts the key time series features, and obtains the time dependency feature set. The specific process is as follows:
[0090] The sequence analysis submodule performs time series analysis on the equipment's operating data through a long short-term memory network based on fault association information, verified data labels, and basic status data sets, collects vibration, temperature, current, and other data of the equipment in different time periods, constructs time series input, and defines the structure and parameters of the long short-term memory network, including the input layer, hidden layer, and output layer. The network optimizes weights and biases through the back propagation algorithm, and the formula is expressed as: "Error = actual value - predicted value". Through iterative training, the network can identify the time dependency and periodic changes of the data, extract key time series features, including the amplitude of periodic changes and time intervals, and the formula is expressed as: "Eigenvalue = maximum value - minimum value / number of cycles", and obtain a time dependency feature set.
[0091] The pattern recognition submodule uses a support vector machine to perform pattern recognition on the data based on the time-dependent feature set, analyze and classify the normal operation mode and multiple abnormal modes of the equipment, and obtain the equipment mode classification results in the following process:
[0092] The pattern recognition submodule performs pattern recognition on the data based on the time-dependent feature set through the support vector machine, standardizes the time-dependent features, defines the classification boundary of the support vector machine, calculates the distance from each feature point to the classification boundary, and divides the equipment operation data into normal operation mode and multiple abnormal modes according to the classification results of the support vector machine. It is mapped to the high-dimensional space through the kernel function, optimizes the classification effect, and obtains the equipment mode classification result.
[0093] Support vector machine, according to the formula:
[0094]
[0095] Calculate the classification results and obtain the device mode classification results, where α i is the Lagrange multiplier corresponding to the i-th data point, indicating the importance of the data point in forming the decision boundary, y i is the class label of the i-th data point, x i is the i-th support vector, which is the key data point on the classification decision boundary, and x is the input feature, which is the data point that needs to be classified. γ is the width parameter of the function, which indicates the relationship between the i-th and j-th data points and controls the kernel function. The decay rate of , e is the logarithm of the natural logarithm, λ is the weight coefficient of the hyperbolic tangent function, which is used to balance the contribution of the hyperbolic tangent kernel, β is the scale parameter of the hyperbolic tangent function, and the control function tanh(β·∥x i -x∥), b is the bias term, n represents the total number of support vectors, and f(x) is the classification function.
[0096] The specific execution process of the formula is:
[0097] According to the input feature x and each support vector x i The distance calculation is based on the similarity of the RBF kernel Then calculate the hyperbolic tangent function tanh(β·∥x i -x∥), the function adds a nonlinear response to the distance to better handle different data distributions, and the result of each function is multiplied by the corresponding α i y i , where α i and i Indicates the importance of data points and their categories, weights the results and adds the bias term b, adjusts λ and β, uses the cross-validation method to test different combinations of λ and β values on the training data, evaluates the classification performance of each set of parameters, selects the λ and β values that minimize the classification error, confirms the λ and β coefficient values, optimizes the contribution ratio and scale of the hyperbolic tangent function, ensures that the final function value is used to determine the classification result through the sign function, and determines the category of the input feature x.
[0098] Based on the equipment pattern classification results, the trend identification submodule uses Transformer technology to analyze long-term data trends and abnormal behavior development, predict the behavior of the equipment in the future, extract the change trend and abnormal pattern of the equipment status, and obtain the abnormal trend analysis results. The specific process is as follows;
[0099] The trend identification submodule is based on the equipment pattern classification results. It uses Transformer technology to analyze long-term data trends and abnormal behavior development, collects long-term equipment operation data, and constructs the input sequence of the Transformer model, which contains data features of multiple time steps. The model calculates the weight of each time step through the self-attention mechanism. The formula is expressed as: "Weight = soft maximum (attention score)". Through the multi-head attention mechanism, the features of different time steps are weighted averaged to extract the change trend of the equipment status. According to the output of the model, the behavior of the equipment in the future period is predicted. The formula is expressed as: "Prediction value = ∑ (weight * feature value) / number of time steps", and abnormal patterns and trends are identified to obtain abnormal trend analysis results.
[0100] See also Figure 2 and Figure 7 ,The fault prediction and warning module includes a state monitoring submodule, a fault analysis submodule, and a warning generation submodule;
[0101] The status monitoring submodule monitors the operating status of lightweight equipment in real time based on the abnormal trend analysis results, evaluates the deviation of data from the normal operating mode, identifies performance degradation and abnormal conditions, and obtains the real-time status information in the following specific processes;
[0102] Based on the abnormal trend analysis results, the status monitoring submodule monitors the operating status of lightweight equipment in real time, collects data such as vibration, temperature, and current during equipment operation, defines the reference value of the normal operating mode, and calculates the deviation between the real-time data and the reference value. The formula is: "Deviation = real-time value - reference value". Each data point is evaluated through the set threshold range to identify performance degradation and abnormal conditions. The specific calculation method is: "Deviation percentage = (deviation / reference value) * 100%". When the deviation exceeds the set threshold, it is marked as abnormal and real-time status information is obtained.
[0103] The fault analysis submodule analyzes potential faults based on real-time status information and predicts the type and time of faults through machine learning models. The specific process of obtaining fault analysis information is as follows:
[0104] The fault analysis submodule analyzes potential faults based on real-time status information, collects and organizes abnormal data in real-time status information, and inputs it into the machine learning model for training. The model predicts the type and time of fault by analyzing historical data. The calculation formula in the prediction process is: "Fault probability = ∑ (weight * abnormal feature value) / total number of features". The weight of each feature value is determined through the model training process, and the possibility and expected time of fault are finally output. Based on the calculation results, the potential fault type and occurrence time are identified to obtain fault analysis information.
[0105] The warning generation submodule adjusts the warning parameters based on the fault analysis information. The warning parameters include the threshold of fault severity and the response time, optimizes the accuracy and response speed of the alarm, and generates abnormal warning information in the following process:
[0106] The early warning generation submodule adjusts the early warning parameters based on the fault analysis information, and sets the threshold of fault severity and response time. The threshold calculation formula for fault severity is: "Severity threshold = baseline value * severity coefficient". The response time is adjusted according to the fault type and historical response time, and the calculation formula is: "Response time = average response time - (prediction time / fault probability)". Optimize the accuracy and response speed of the alarm, apply the optimized early warning parameters to the alarm system, and generate abnormal early warning information.
[0107] See also Figure 2 and Figure 8 ,The maintenance optimization decision module includes the maintenance demand analysis submodule, the maintenance priority evaluation submodule, and the maintenance strategy adjustment submodule;
[0108] The maintenance demand analysis submodule analyzes the maintenance demand of lightweight equipment based on abnormal warning information, combined with the historical maintenance records and fault data of lightweight equipment, identifies key maintenance points and the urgency of maintenance, and formulates a maintenance task list. The specific process of obtaining the maintenance demand analysis results is as follows;
[0109] The maintenance demand analysis submodule analyzes the maintenance demand of lightweight equipment based on abnormal warning information, combined with the historical maintenance records and fault data of lightweight equipment, collects the historical maintenance records and fault data of the equipment, and counts the frequency and effect of various maintenance activities. It uses abnormal warning information to identify the key maintenance points of the current equipment and calculates the urgency of each maintenance point. The formula is: "Emergency = Warning level * Fault frequency / Historical maintenance effect". According to the calculation results, the key points that need immediate maintenance are identified first, and a detailed maintenance task list is formulated to obtain the maintenance demand analysis results.
[0110] The maintenance priority assessment submodule uses a multi-criteria decision analysis method based on the maintenance demand analysis results, combined with the potential impact of failures, equipment criticality and operation interruption costs, to evaluate the priorities of multiple maintenance activities and optimize the continuous operation of key equipment. The specific process of obtaining maintenance priority information is as follows;
[0111] The maintenance priority assessment submodule is based on the maintenance needs analysis results and uses a multi-criteria decision analysis method to combine the potential impact of failures, equipment criticality, and operation interruption costs to evaluate the priorities of multiple maintenance activities, define the weight of each criterion, calculate the comprehensive score of each maintenance activity, sort the comprehensive scores, identify the maintenance activities with the highest priority, ensure the continuous operation of key equipment, and obtain maintenance priority information.
[0112] Multi-criteria decision analysis method, according to the formula:
[0113]
[0114] Calculate the priority score of the maintenance activity and obtain the maintenance priority information, where:
[0115] S i is the total score of the ith maintenance activity, w j is the weight of the jth criterion, c ij is the score of the ith maintenance activity under the jth criterion, m is the total number of criteria, indicating the number of different aspects or criteria considered in the evaluation process, k is the slope parameter of the curve, which is used to adjust the sensitivity of the function to the difference between the score and the threshold, τ j is the score threshold under the jth criterion, and e is the base of the natural logarithm.
[0116] The specific execution process of the formula is as follows:
[0117] For each maintenance activity i, calculate the score c for each evaluation criterion j ij The score is calculated based on whether the respective threshold τ is reached. j , adjusted by the response function, using the formula To enhance or weaken the impact of the score. The adjusted score is related to the corresponding weight w j Multiply them together and then sum all the criteria to form the total score of the first part. In order to increase the sensitivity to the variability of the scores and the response to outliers, the score of the second part is calculated by summing the square of the weights of all the criteria and the logarithm of the scores and taking the square root. The two parts of the score are added to get the final priority score S for each maintenance activity. i .
[0118] The maintenance strategy adjustment submodule adjusts the time and resource allocation of the maintenance plan based on the maintenance priority information, the current operating status and maintenance cost of the lightweight equipment, and optimizes the cost efficiency of maintenance. The specific process of obtaining maintenance decision information is as follows;
[0119] The maintenance strategy adjustment submodule adjusts the time and resource allocation of the maintenance plan based on the maintenance priority information, the current operating status of the lightweight equipment and the maintenance cost, evaluates the current operating status of the equipment, and calculates the cost efficiency of the maintenance plan. The formula is: "cost efficiency = maintenance cost / maintenance effect". Then, according to the operating status and priority information of the equipment, the maintenance resources are reallocated and the execution time of the maintenance plan is adjusted to ensure the best maintenance effect at the lowest cost and obtain maintenance decision information.
[0120] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.
Claims
1. A lightweight equipment fault prediction system based on multidimensional data, Features: The system comprises: The data acquisition module captures the operating status data of the lightweight device based on the operating environment of the lightweight device, and performs timestamp marking and format standardization on the data to obtain a basic status data set; The data label processing module processes the unlabeled data based on the basic state data set by using a semi-supervised learning method, and performs a consistency check on the labels to obtain verified data labels; The fault association identification module uses a graph neural network to model the component functions and failure modes of the lightweight equipment based on the verified data labels, captures the propagation path of the fault, and obtains fault association information; The abnormal trend analysis module analyzes the operation data of the lightweight equipment using Transformer and long short-term memory network based on the fault association information, verified data labels and basic status data set, extracts the trend changes and abnormal patterns of the data, and obtains the abnormal trend analysis results; The fault prediction and warning module predicts the time and type of lightweight equipment failure based on the abnormal trend analysis results, adjusts the alarm parameters, optimizes the accuracy and response speed of the alarm, and obtains abnormal warning information; The maintenance optimization decision module evaluates the current maintenance requirements of the equipment, evaluates the priority of maintenance activities, adjusts the maintenance sequence and maintenance time, and obtains maintenance decision information based on the abnormal warning information.
2. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The basic status data set includes time synchronization data, format unified data and noise filtered data, the verified data labels include normal labels, abnormal labels and misjudgment labels, the fault association information includes graphical equipment structure, equipment component dependencies and predicted functional failure points, the abnormal trend analysis results include time series trend data, abnormal mode indicators and potential fault signals, the abnormal warning information includes status monitoring data, fault mode identification and warning signal information, and the maintenance decision information includes cost-effectiveness information, maintenance strategy information and maintenance time points.
3. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The data acquisition module comprises: The time stamp submodule captures the vibration, temperature and current data of the lightweight equipment based on its operating environment, performs time stamping, and generates time series data for each data record corresponding to the date and time. The format standardization submodule performs data format unification processing based on the time series data, converts the multi-source data into a unified data format, and obtains unified format data; The noise filtering submodule performs noise filtering on the data stream based on the unified format data, removes electrical interference and atypical data noise points through filtering, optimizes the accuracy and reliability of the data, and obtains a basic status data set.
4. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The data label processing module includes: The data classification submodule performs machine learning classification on the unlabeled data based on the basic state data set, identifies patterns and categories in the data, and obtains category identification data; The labeling processing submodule uses a semi-supervised learning method to label the data based on the category recognition data, assigns corresponding labels to each data point, including normal, abnormal and warning, and converts them into a machine-readable form through encoding technology to optimize the consistency and operability of the data set to obtain processed label data; The consistency check submodule performs consistency check on the label based on the processed label data, adopts statistical methods and logic verification to test the consistency and accuracy of the data label, verifies that the label reflects the state and characteristics of the data, and obtains the verified data label.
5. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The fault association identification module comprises: The relationship analysis submodule analyzes the functional associations between various components of the lightweight equipment based on the verified data tags, identifies the interactions and dependencies between component functions through statistical correlation analysis of the data, and obtains component function mapping information; The performance index extraction submodule identifies and extracts key performance indicators based on the component function mapping information, including abnormal temperature and current mutation associated with the fault, and identifies the performance indicators associated with the fault mode by evaluating the performance indicators, thereby obtaining key performance indicator information; Based on the key performance indicator information, the dependency building submodule constructs a dependency model between lightweight device components through graph neural network technology, and obtains fault correlation information by adjusting the weights of nodes and edges to identify the impact and fault propagation path between components.
6. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The abnormal trend analysis module includes: The sequence analysis submodule performs time series analysis on the operation data of the equipment through a long short-term memory network based on the fault association information, the verified data labels and the basic state data set, identifies the time dependency and periodic changes of the data, extracts key time series features, and obtains a time dependency feature set; The pattern recognition submodule performs pattern recognition on the data based on the time-dependent feature set through a support vector machine, analyzes and classifies the normal operation mode and multiple abnormal modes of the equipment, and obtains the equipment mode classification result; Based on the equipment mode classification results, the trend identification submodule analyzes long-term data trends and abnormal behavior development through Transformer technology, predicts the behavior of the equipment in the future, extracts the change trend and abnormal pattern of the equipment status, and obtains the abnormal trend analysis results.
7. The lightweight equipment failure prediction system based on multidimensional data according to claim 6, Features: The support vector machine is according to the formula: Calculate the classification results and obtain the device mode classification results, where α i is the Lagrange multiplier, y i is the class label of the i-th data point, x i is the i-th support vector, x is the input feature, γ is the width parameter of the function, e is the logarithm of the natural logarithm, λ is the weight coefficient of the hyperbolic tangent function, β is the scale parameter of the hyperbolic tangent function, b is the bias term, n represents the total number of support vectors, and f(x) is the classification function.
8. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The fault prediction and warning module comprises: The state monitoring submodule monitors the operating state of the lightweight equipment in real time based on the abnormal trend analysis results, evaluates the deviation of the data from the normal operating mode, identifies performance degradation and abnormal conditions, and obtains real-time state information; The fault analysis submodule analyzes potential faults based on the real-time status information, predicts the type and occurrence time of the fault through a machine learning model, and obtains fault analysis information; The warning generation submodule adjusts warning parameters based on the fault analysis information, the warning parameters including the threshold of fault severity and the response time, optimizes the accuracy and response speed of the alarm, and generates abnormal warning information.
9. The lightweight equipment failure prediction system based on multidimensional data according to claim 1, Features: The maintenance optimization decision module includes: The maintenance demand analysis submodule analyzes the maintenance demand of lightweight equipment based on the abnormal warning information, combined with the historical maintenance records and fault data of lightweight equipment, identifies key maintenance points and the urgency of maintenance, formulates a maintenance task list, and obtains the maintenance demand analysis results; The maintenance priority assessment submodule uses a multi-criteria decision analysis method based on the maintenance demand analysis results, combines the potential impact of the failure, equipment criticality and operation interruption cost, evaluates the priority of multiple maintenance activities, optimizes the continuous operation of key equipment, and obtains maintenance priority information; The maintenance strategy adjustment submodule adjusts the time and resource allocation of the maintenance plan based on the maintenance priority information, according to the current operating status and maintenance cost of the lightweight equipment, optimizes the cost efficiency of maintenance, and obtains maintenance decision information.
10. The lightweight equipment failure prediction system based on multidimensional data according to claim 9, Features: The multi-criteria decision analysis method is based on the formula: Calculate the priority score of the maintenance activity and obtain the maintenance priority information, where: S i is the total score of the ith maintenance activity, w j is the weight of the jth criterion, c ij is the score of the ith maintenance activity under the jth criterion, m is the total number of criteria, k is the slope parameter of the curve, τ j is the score threshold under the jth criterion, and e is the base of the natural logarithm.
Citation Information
Cited By
AI-based science and technology equipment risk intelligent assessment system and method
CN119204709A
An AI-based intelligent risk assessment system and method for scientific and technological equipment
CN119204709B
Christmas tree fault propagation path development trajectory prediction method based on deep learning
CN119226746A
Internet-of-things five-primary-color full-spectrum multi-color-temperature LED peak shifting energy storage intelligent street lamp digital platform
CN119421306A
Intelligent cell gate fault early warning system based on combinatorial algorithm
CN119474886A
