An elevator hoisting machine health assessment method and system based on multi-source data fusion

Abstract

The present application relates to the technical field of elevator health, and discloses a kind of elevator hoisting machine health evaluation method and system based on multi-source data fusion, the method comprises: extracting the time domain feature and frequency domain feature of multi-source sensor data in elevator hoisting machine, obtain the feature set of the elevator hoisting machine;Determine the entropy value of each feature in the feature set;According to the information uncertainty of the feature represented by the entropy value, the feature weight of the feature is assigned, and the feature is weighted and fused according to the feature weight, to obtain the fusion feature of the elevator hoisting machine;Based on historical elevator hoisting machine data, the health evaluation model of the elevator hoisting machine is constructed;The fusion feature is input into the health evaluation model to generate the health index of the elevator hoisting machine;According to the health index, the elevator hoisting machine is maintained and evaluated;The present application can improve the efficiency of elevator hoisting machine health evaluation.

Description

Technical Field

[0001] This invention relates to the field of elevator health technology, and in particular to a method and system for health assessment of elevator traction machines based on multi-source data fusion. Background Technology

[0002] Existing technologies lack a comprehensive signal integrity verification process for processing multi-source sensor data of elevator traction machines, which can easily lead to residual noise and missing key operational information in the data. Furthermore, in the feature extraction stage, time-domain features or frequency-domain features are often extracted separately without being organically combined. The resulting feature set cannot fully cover the key dimensions of the traction machine's operating status, causing subsequent health assessments to lack accurate data support and making it difficult to effectively identify potential subtle anomalies in the equipment, thus affecting the reliability of the assessment results.

[0003] Existing technologies, when determining feature weights, fail to scientifically allocate them based on the uncertainty of feature information reflected by entropy values, relying heavily on manual experience to set weights. This weakens the role of important features and amplifies the influence of secondary features. Furthermore, when constructing health assessment models, the historical elevator traction machine data is not thoroughly cleaned, the operating conditions are not finely divided, and the quality of model training samples is poor. This makes it impossible to accurately establish the mapping relationship between health status and features, resulting in a large deviation between the output health index and the actual health status of the equipment. Consequently, maintenance assessments lack specificity and fail to guide efficient equipment maintenance. Therefore, improving the efficiency of elevator traction machine health assessment has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for health assessment of elevator traction machines based on multi-source data fusion, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for health assessment of elevator traction machines based on multi-source data fusion, comprising:

[0006] S1. Extract the time-domain and frequency-domain features of the multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine;

[0007] S2. Determine the entropy value of each feature in the feature set;

[0008] S3. Assign feature weights to the features based on the information uncertainty represented by the entropy value, and perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine;

[0009] S4. Based on historical elevator traction machine data, construct a health assessment model for the elevator traction machine;

[0010] S5. Input the fused features into the health assessment model to generate the health index of the elevator traction machine;

[0011] S6. Perform a maintenance assessment on the elevator traction machine based on the health index.

[0012] In a preferred embodiment, the step of extracting the time-domain and frequency-domain features of the multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine includes:

[0013] Signal integrity verification is performed on the multi-source sensor data of the elevator traction machine to obtain the sensor data sequence of the elevator traction machine;

[0014] Based on the statistical mean, variance, and peak factor of the sensor data sequence, a time-domain feature subset of the elevator traction machine is generated;

[0015] The frequency domain transformation of the sensor data sequence is performed to obtain the frequency domain feature subset of the elevator traction machine;

[0016] The time-domain feature subset and the frequency-domain feature subset are combined to generate the feature set of the elevator traction machine.

[0017] In a preferred embodiment, determining the entropy value of each feature in the feature set includes:

[0018] The feature set is segmented to obtain the feature data subsequence of the feature set;

[0019] The feature data subsequence is divided into states to obtain the state sequence corresponding to the feature data subsequence;

[0020] The approximate entropy of each feature in the feature set is calculated based on the state transition probability of the state sequence to obtain the entropy value of the elevator traction machine.

[0021] In a preferred embodiment, assigning feature weights to the features based on the information uncertainty of the features represented by the entropy value includes:

[0022] Based on the information uncertainty of the entropy value representing the feature, the weight coefficients of the feature are assigned to obtain the initial weight allocation scheme of the feature;

[0023] Perform a weight consistency check on the initial weight allocation scheme to obtain the verified feature weights of the features;

[0024] The verified feature weights are smoothed to obtain the feature weight distribution of the feature;

[0025] The features are ranked in order of importance according to the feature weight distribution, and the feature weights are...

[0026] In a preferred embodiment, the step of assigning weight coefficients to the features based on the information uncertainty of the entropy value to obtain an initial weight allocation scheme for the features includes:

[0027] The entropy value is divided into multiple uncertainty levels and assigned corresponding benchmark weight coefficients;

[0028] Based on the correlation constraints between features, the baseline weight coefficients are collaboratively optimized to obtain the collaborative weight coefficients of the features.

[0029] The collaborative weight coefficient is dynamically adjusted based on the accuracy of the historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature.

[0030] In a preferred embodiment, the step of dynamically adjusting the collaborative weight coefficient based on the accuracy of historical evaluation results in the elevator traction machine to obtain an initial weight allocation scheme for the feature includes:

[0031] Collect health index output results and actual maintenance records within historical assessment periods to obtain a sample set of assessment accuracy for the aforementioned features;

[0032] By performing correlation analysis between the evaluation accuracy sample set and the corresponding collaborative weight coefficients, the weight configuration patterns that cause evaluation bias can be identified.

[0033] Based on the mapping relationship between accuracy and weight adjustment amount, the weight correction parameters of the weight configuration mode are generated;

[0034] When the accuracy of multiple consecutive evaluation cycles reaches a stable threshold, the weight configuration of the current weight configuration mode is locked to obtain the initial weight allocation scheme of the feature.

[0035] In a preferred embodiment, constructing a health assessment model for the elevator traction machine based on historical elevator traction machine data includes:

[0036] Data cleaning and operating condition classification were performed on historical elevator traction machine data to obtain the training sample set of the elevator traction machine.

[0037] Multi-scale feature extraction is performed on the training sample set to obtain the feature representation space of the elevator traction machine;

[0038] The features in the feature representation space are ranked by importance to obtain the feature subset of the elevator traction machine;

[0039] By performing layer-by-layer feature abstraction on the feature subset, the health status mapping relationship of the elevator traction machine is obtained;

[0040] A hierarchical evaluation network for the elevator traction machine is constructed based on the health status mapping relationship;

[0041] The hierarchical evaluation network is trained and optimized to form the health assessment model for the elevator traction machine.

[0042] In a preferred embodiment, the step of training and optimizing the hierarchical evaluation network, and the health assessment model of the elevator traction machine, includes:

[0043] Adversarial training is performed on the hierarchical evaluation network to obtain the initial training model of the hierarchical evaluation network;

[0044] Identify the key feature layers in the initial training model and adjust the network connection structure of the initial training model to obtain the structure optimization model of the hierarchical evaluation network;

[0045] Based on the changing trend of the training loss curve of the adversarial training, the parameter update step size of the structure optimization model is adaptively adjusted to obtain the parameter optimization model of the hierarchical evaluation network.

[0046] The evaluation capabilities of multiple expert models are transferred to the parameter optimization model to obtain the health assessment model of the elevator traction machine.

[0047] In a preferred embodiment, the maintenance assessment of the elevator traction machine based on the health index includes:

[0048] Establish a rule base for mapping historical health indices to maintenance strategies;

[0049] Preliminary maintenance recommendations for the elevator traction machine are generated based on the numerical range of the health index.

[0050] Based on the operating environment parameters of the elevator traction machine, the preliminary maintenance recommendations are modified to adapt to the operating conditions, resulting in an environmentally adapted maintenance scheme for the elevator traction machine.

[0051] Based on the aforementioned environment-adaptive maintenance scheme, the execution priority and resource allocation requirements of the maintenance procedures are clearly defined;

[0052] The elevator traction machine is maintained according to the execution priority and resource configuration requirements.

[0053] To address the aforementioned problems, this invention also provides an elevator traction machine health assessment system based on multi-source data fusion, the system comprising:

[0054] A multi-source feature extraction module is used to extract the time-domain and frequency-domain features of multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine;

[0055] The feature entropy value determination module is used to determine the entropy value of each feature in the feature set;

[0056] The entropy weight fusion feature construction module is used to assign feature weights to the features based on the information uncertainty of the features represented by the entropy value, and to perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine.

[0057] The health assessment model construction module is used to construct a health assessment model for the elevator traction machine based on historical elevator traction machine data.

[0058] A health index generation module is used to input the fused features into the health assessment model to generate the health index of the elevator traction machine;

[0059] The maintenance assessment and decision-making module is used to assess the maintenance of the elevator traction machine based on the health index.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. This invention verifies the signal integrity of multi-source sensor data from elevator traction machines, extracts the mean, variance, and peak factor of the data to generate a time-domain feature subset, and then performs frequency domain transformation to obtain a frequency-domain feature subset. The two are combined to form a comprehensive feature set. At the same time, entropy is used to characterize the uncertainty of feature information, and feature weights are scientifically allocated and weighted to obtain fused features. This ensures that the features cover the key dimensions of equipment operation and that the weights are appropriate to the information value, providing accurate and reliable feature support for health assessment.

[0062] 2. This invention cleans and classifies historical elevator traction machine data to construct a training sample set. Through multi-scale feature extraction and hierarchical network training optimization, a health assessment model is obtained. Input fusion features generate an accurate health index. Then, combined with the historical mapping rule base and operating environment parameters, an adapted maintenance plan is generated, clarifying the priority of maintenance procedures and resource allocation, thereby achieving accurate health assessment and efficient maintenance decision-making, effectively ensuring the stable operation of the elevator traction machine. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating a method for assessing the health of an elevator traction machine based on multi-source data fusion, provided in an embodiment of the present invention.

[0064] Figure 2 A functional block diagram of an elevator traction machine health assessment system based on multi-source data fusion provided in an embodiment of the present invention;

[0065] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0066] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0067] This application provides a method for health assessment of elevator traction machines based on multi-source data fusion. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for health assessment of elevator traction machines based on multi-source data fusion can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0068] Reference Figure 1 The diagram shown is a flowchart illustrating a method for assessing the health of an elevator traction machine based on multi-source data fusion, according to an embodiment of the present invention. In this embodiment, the method for assessing the health of an elevator traction machine based on multi-source data fusion includes:

[0069] S1. Extract the time-domain and frequency-domain features of the multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine;

[0070] In this embodiment of the invention, the step of extracting the time-domain and frequency-domain features of multi-source sensor data from the elevator traction machine to obtain the feature set of the elevator traction machine includes:

[0071] Signal integrity verification is performed on the multi-source sensor data of the elevator traction machine to obtain the sensor data sequence of the elevator traction machine;

[0072] Based on the statistical mean, variance, and peak factor of the sensor data sequence, a time-domain feature subset of the elevator traction machine is generated;

[0073] The frequency domain transformation of the sensor data sequence is performed to obtain the frequency domain feature subset of the elevator traction machine;

[0074] The time-domain feature subset and the frequency-domain feature subset are combined to generate the feature set of the elevator traction machine.

[0075] Before processing the multi-source sensor data of the elevator traction machine, the types of multi-source sensors must be identified, including vibration sensors that detect traction machine vibration, temperature sensors that monitor operating temperature, and speed sensors that record rotational speed. Raw data transmitted in real time from each sensor must be collected. When performing signal integrity verification, data continuity is checked first. If three or more consecutive data points are missing from a sensor, linear interpolation using five normal data points before and after the missing period is used to fill in the missing values. Next, the data rationality is checked by comparing each data point with the corresponding sensor's range. For example, if the vibration sensor's range is 0-50 mm / s², data exceeding this range is considered an anomaly and replaced with the average data from the same sensor under the same operating conditions for that day. After verification, the valid data from each sensor are integrated in chronological order to form an ordered sequence of elevator traction machine sensor data.

[0076] Extract target data segments within the same time interval from the sensor data sequence. Calculate the characteristic mean of this data segment: sum all data values ​​within the segment, divide the sum by the total number of data points in the segment, and obtain the characteristic mean. Calculate the variance: subtract the characteristic mean from each data value within the segment to obtain the deviation value for each data point. Square all deviation values, sum the squared results, and divide the sum by the total number of data points to obtain the variance. Calculate the peak factor: first find the maximum data value within the segment, then calculate the root mean square of the segment, and divide the peak value by the root mean square to obtain the peak factor. Organize the characteristic mean, variance, and peak factor of the same data segment into a set of characteristic data. Generate multiple sets of characteristic data sequentially according to sensor type and time interval, combining them to form a time-domain characteristic subset of the elevator traction machine.

[0077] The sensor data sequence is divided into multiple independent data blocks at fixed time intervals, ensuring that each data block has the same length. A frequency domain transformation is performed on each data block: first, the time-domain data within the data block is converted to frequency-domain data. Through this transformation, the amplitude values ​​corresponding to different frequencies of the data block are obtained. Key frequency domain features are extracted from the transformation results: the maximum amplitude value for each frequency is identified, and the frequency corresponding to this maximum amplitude is recorded. Simultaneously, the top three frequencies with the highest amplitude values ​​and their corresponding amplitude values ​​are selected. The maximum amplitude, primary frequency, secondary frequency, and corresponding amplitude values ​​of each data block are compiled into a set of frequency domain feature data. The frequency domain feature data of all data blocks are then integrated to form a subset of the frequency domain features of the elevator traction machine.

[0078] All feature data within the time-domain feature subset are labeled in the format of "sensor type-time interval-feature name", such as "vibration sensor-08:00-09:00-mean value"; similarly, feature data within the frequency-domain feature subset are labeled in the format of "sensor type-data block number-feature name", such as "speed sensor-5th data block-main frequency". Classified by sensor type, time-domain and frequency-domain feature data corresponding to the same sensor are grouped together. The time-domain feature data is arranged first, followed by the corresponding frequency-domain feature data, ensuring a one-to-one correspondence between time-domain and frequency-domain features within the same time dimension. All categorized and organized feature data are integrated into a complete feature set to generate the feature set for the elevator traction machine.

[0079] The beneficial effects are as follows: signal integrity verification removes abnormal data and supplements missing data, ensuring the accuracy and continuity of sensor data sequences and providing a reliable data foundation for subsequent feature extraction; by calculating the feature mean, variance, and peak factor to generate a time-domain feature subset, it can reflect the overall distribution and fluctuation characteristics of the data; the frequency-domain feature subset obtained by frequency domain transformation can capture key information in the frequency domain of the data; the feature set formed by the combination of the two comprehensively covers the time-domain and frequency-domain characteristics of the traction machine's operating status, providing comprehensive and accurate feature support for subsequent determination of feature entropy values, weight allocation, and construction of health assessment models, effectively improving the quality of preliminary data processing for elevator traction machine health assessment.

[0080] S2. Determine the entropy value of each feature in the feature set;

[0081] In this embodiment of the invention, determining the entropy value of each feature in the feature set includes:

[0082] The feature set is segmented to obtain the feature data subsequence of the feature set;

[0083] The feature data subsequence is divided into states to obtain the state sequence corresponding to the feature data subsequence;

[0084] The approximate entropy of each feature in the feature set is calculated based on the state transition probability of the state sequence to obtain the entropy value of the elevator traction machine.

[0085] First, define the structure of the feature set, which contains feature data of various types arranged in ascending order of timestamps, and set a fixed segmentation window. Starting from the first data point in the feature set, sequentially extract 60 consecutive data points in chronological order to form a subsequence. After extraction, skip the data points already included in the subsequence and continue extracting the next subsequence containing 60 consecutive data points, until all data points in the feature set are covered. If there are fewer than 60 data points remaining, merge the remaining data points with the last data point of the previous subsequence to ensure that the data volume of each subsequence is consistent, ultimately obtaining the feature data subsequences of the feature set.

[0086] First, calculate the numerical range of each feature data subsequence. Divide this range into three equally spaced intervals, defining three states: "normal state" (A) where the value falls within the smallest interval, "state of interest" (B) where it falls within the middle interval, and "abnormal state" (C) where it falls within the largest interval. Extract each data point from the feature data subsequence one by one, compare the value of each data point with the boundary values ​​of the three intervals to determine the state label corresponding to that data point. Arrange all state labels into an ordered character sequence according to the order in which the data points appear in the subsequence, thus obtaining the state sequence corresponding to the feature data subsequence.

[0087] First, count the total number of occurrences of each state in the state sequence and the number of transitions between different states. For example, traverse the state sequence and record the total number of occurrences of "normal state," "attention state," and "abnormal state," while also recording the number of occurrences of the nine state transitions: A→A, A→B, A→C, B→A, B→B, B→C, C→A, C→B, and C→C. Divide the number of each state transition by the total number of occurrences of the initial state for that transition to obtain the corresponding state transition probability. Next, calculate the approximate entropy: set the similarity tolerance to 0.15 times the standard deviation of the feature data subsequence, split the state sequence into multiple continuous vectors of length m=2, calculate the similarity of each vector to all other vectors, count the proportion of similar vectors to the total number of vectors, take the natural logarithm of each proportion, and calculate the average to obtain the average logarithm value when m=2. Then set m to 3 and repeat the above steps of vector splitting, similarity calculation, and average logarithm calculation to obtain the average logarithm value when m=3. The approximate entropy of a feature is obtained by subtracting the average logarithm of m=3 from the average logarithm of m=2. The approximate entropies of all features together constitute the entropy value of the elevator traction machine.

[0088] The beneficial effects are as follows: by segmenting the feature set through a fixed window to obtain feature data subsequences, the subsequences can accurately reflect the operating characteristics of the elevator traction machine during a specific time period, providing timely basic data for subsequent state analysis; by dividing the state according to the numerical range to obtain the state sequence, the continuous feature data is transformed into quantifiable discrete states, clearly presenting the changing patterns of the operating state; and by calculating the approximate entropy based on the state transition probability to obtain the entropy value, the information uncertainty of each feature can be objectively quantified, providing a reliable basis for the subsequent scientific allocation of feature weights based on the entropy value, effectively improving the accuracy and objectivity of feature analysis in the health assessment of the elevator traction machine.

[0089] S3. Assign feature weights to the features based on the information uncertainty represented by the entropy value, and perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine;

[0090] In this embodiment of the invention, assigning feature weights to the features based on the information uncertainty of the features represented by the entropy value includes:

[0091] Based on the information uncertainty of the entropy value representing the feature, the weight coefficients of the feature are assigned to obtain the initial weight allocation scheme of the feature;

[0092] Perform a weight consistency check on the initial weight allocation scheme to obtain the verified feature weights of the features;

[0093] The verified feature weights are smoothed to obtain the feature weight distribution of the feature;

[0094] The features are ranked in order of importance according to the feature weight distribution, and the feature weights are...

[0095] The step of assigning weight coefficients to the features based on the information uncertainty represented by the entropy value, to obtain the initial weight allocation scheme for the features, includes:

[0096] The entropy value is divided into multiple uncertainty levels and assigned corresponding benchmark weight coefficients;

[0097] Based on the correlation constraints between features, the baseline weight coefficients are collaboratively optimized to obtain the collaborative weight coefficients of the features.

[0098] The collaborative weight coefficient is dynamically adjusted based on the accuracy of the historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature.

[0099] The method of dynamically adjusting the collaborative weight coefficient based on the accuracy of historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature includes:

[0100] Collect health index output results and actual maintenance records within historical assessment periods to obtain a sample set of assessment accuracy for the aforementioned features;

[0101] By performing correlation analysis between the evaluation accuracy sample set and the corresponding collaborative weight coefficients, the weight configuration patterns that cause evaluation bias can be identified.

[0102] Based on the mapping relationship between accuracy and weight adjustment amount, the weight correction parameters of the weight configuration mode are generated;

[0103] When the accuracy of multiple consecutive evaluation cycles reaches a stable threshold, the weight configuration of the current weight configuration mode is locked to obtain the initial weight allocation scheme of the feature.

[0104] Entropy values ​​are categorized into three uncertainty levels based on their numerical magnitude: low uncertainty (0-0.3, indicating high reliability of feature information, assigned a baseline weight coefficient of 0.8); medium uncertainty (0.3-0.6, indicating moderate reliability of feature information, assigned a baseline weight coefficient of 0.5); and high uncertainty (above 0.6, indicating low reliability of feature information, assigned a baseline weight coefficient of 0.2). For each feature in the feature set, interval matching is performed to determine its uncertainty level, and a corresponding baseline weight coefficient is assigned, thus completing the baseline weight coefficient allocation for all features.

[0105] First, correlation constraints are determined by calculating the numerical similarity between features. If the numerical similarity between two features exceeds 80%, they are considered strongly correlated features. The constraint is that the sum of the baseline weight coefficients of strongly correlated features must not exceed 1.2, and the difference between their weights must be less than 0.3. All features are traversed to identify strongly correlated feature combinations. The baseline weight coefficients of features within these combinations are then co-optimized: if the baseline weight coefficients of two features in a strongly correlated combination are 0.8 and 0.7 respectively, and the sum of 1.5 exceeds the constraint, they are proportionally reduced to 0.72 and 0.63 to ensure compliance with the constraints. The baseline weight coefficients of non-strongly correlated features remain unchanged. Finally, the co-weight coefficients of all features are obtained.

[0106] Health index outputs from the past 12 assessment periods were collected, along with actual maintenance records of the elevator traction machine for each period, including whether a fault occurred, the type of fault, and the time of occurrence. The health index outputs for each period were compared with the actual maintenance records: if the health index was determined to be "normal" and there was no actual fault, or if the health index was determined to be "abnormal" and a corresponding fault occurred, it was marked as "assessment accurate"; if the health index did not match the actual situation, it was marked as "assessment biased." Following the assessment period sequence, the "assessment accuracy / biasedness" results for each period were compiled and correlated with the collaborative weighting coefficients used in that period, forming a characteristic assessment accuracy sample set.

[0107] The samples exhibiting "assessment bias" in the accuracy assessment sample set are extracted separately and correlated one-to-one with their corresponding collaborative weight coefficients. The common characteristics of the collaborative weight coefficients in the biased samples are analyzed. If multiple groups of biased samples are found to have "collaborative weight coefficients for low-uncertainty features below 0.6 and collaborative weight coefficients for high-uncertainty features above 0.4," this weight configuration is determined to be a weight configuration pattern leading to assessment bias. If the biased samples share the common characteristic of "the difference in weights for strongly correlated features exceeding 0.5," this configuration is also classified as a biased weight configuration pattern, thus completing the identification of all biased weight configuration patterns.

[0108] A mapping table between accuracy and weight adjustment is pre-established: when the evaluation accuracy is between 70% and 80%, the co-weight coefficient of low-uncertainty features increases by 0.1, and the co-weight coefficient of high-uncertainty features decreases by 0.05; when the evaluation accuracy is between 60% and 70%, the co-weight coefficient of low-uncertainty features increases by 0.15, and the co-weight coefficient of high-uncertainty features decreases by 0.1; when the evaluation accuracy is below 60%, the co-weight coefficient of low-uncertainty features increases by 0.2, and the co-weight coefficient of high-uncertainty features decreases by 0.15. Based on the evaluation accuracy corresponding to the identified bias weight configuration pattern, the matching adjustment amount is found in the mapping table to generate the weight correction parameters for that weight configuration pattern.

[0109] A stability threshold is set at which the evaluation accuracy reaches 90% or higher for three consecutive evaluation periods. The accuracy is monitored chronologically after the weight adjustment parameters are applied. After each evaluation period, the evaluation accuracy is calculated and compared to the stability threshold. If the accuracy is 92% for the first period, 91% for the second, and 93% for the third, meeting the stability threshold requirement for three consecutive periods, the current weight configuration mode is immediately locked. The weight coefficients of all features under this configuration mode are then compiled into a structured document to obtain the initial weight allocation scheme for the features.

[0110] Two consistency checks are performed on all feature weight coefficients in the initial weight allocation scheme: The first check verifies whether the sum of all feature weight coefficients is within the range of 0.95-1.05. If the sum is 1.12, exceeding the upper limit, the sum is reduced proportionally to the weight coefficient of each feature, adjusting the total to 1.0. The second check verifies whether the weight coefficient of an individual feature is within the range of 0-1. If a feature weight coefficient is 1.05, exceeding the upper limit, it is directly adjusted to 1.0; if a feature weight coefficient is -0.02, below the lower limit, it is adjusted to 0.02. After completing these two checks, the verified feature weights for all features are obtained.

[0111] The adjacent feature weight smoothing method is used to process the verified feature weights. The average of the weight coefficients of each feature and its two adjacent features is calculated, and this average is used as the smoothed weight coefficient for that feature. For example, if feature A has a weight of 0.9, feature B has a weight of 0.1, and feature C has a weight of 0.8, the smoothed weight of feature B is (0.9 + 0.1 + 0.8) / 3 ≈ 0.6, the smoothed weight of feature A is (0.1 + 0.9 + 0.9) / 3 ≈ 0.63, and the smoothed weight of feature C is (0.1 + 0.8 + 0.8) / 3 ≈ 0.57. The smoothed weight coefficients of all features are then arranged in the original feature order to obtain the feature weight distribution.

[0112] Based on the weight coefficients of each feature in the feature weight distribution, the features are arranged in descending order, with the feature with the largest weight coefficient at the top, decreasing sequentially, and the feature with the smallest weight coefficient at the bottom. During the sorting process, if two features have the same weight coefficient, they are sorted by their corresponding entropy values ​​from smallest to largest. After sorting, each feature is matched with its corresponding weight coefficient and recorded to obtain the feature weight.

[0113] The beneficial effects are as follows: by assigning benchmark weights through entropy value grading, optimizing weights by combining correlation constraints, verifying and adjusting weights based on historical data, ensuring the compliance of weights through consistency verification, smoothing out weight abrupt changes, and finally clarifying the importance of features through ranking, the entire process is progressive. It not only ensures the adaptability of feature weights to information uncertainty, but also ensures the consistency, stability and rationality of weights. This provides a reliable basis for subsequent feature weighted fusion based on weights, effectively improves the quality of elevator traction machine fusion features, and thus enhances the accuracy of health assessment results.

[0114] S4. Based on historical elevator traction machine data, construct a health assessment model for the elevator traction machine;

[0115] In this embodiment of the invention, constructing a health assessment model for the elevator traction machine based on historical elevator traction machine data includes:

[0116] Data cleaning and operating condition classification were performed on historical elevator traction machine data to obtain the training sample set of the elevator traction machine.

[0117] Multi-scale feature extraction is performed on the training sample set to obtain the feature representation space of the elevator traction machine;

[0118] The features in the feature representation space are ranked by importance to obtain the feature subset of the elevator traction machine;

[0119] By performing layer-by-layer feature abstraction on the feature subset, the health status mapping relationship of the elevator traction machine is obtained;

[0120] A hierarchical evaluation network for the elevator traction machine is constructed based on the health status mapping relationship;

[0121] The hierarchical evaluation network is trained and optimized to form the health assessment model for the elevator traction machine.

[0122] The training and optimization of the hierarchical evaluation network, and the resulting health assessment model for the elevator traction machine, include:

[0123] Adversarial training is performed on the hierarchical evaluation network to obtain the initial training model of the hierarchical evaluation network;

[0124] Identify the key feature layers in the initial training model and adjust the network connection structure of the initial training model to obtain the structure optimization model of the hierarchical evaluation network;

[0125] Based on the changing trend of the training loss curve of the adversarial training, the parameter update step size of the structure optimization model is adaptively adjusted to obtain the parameter optimization model of the hierarchical evaluation network.

[0126] The evaluation capabilities of multiple expert models are transferred to the parameter optimization model to obtain the health assessment model of the elevator traction machine.

[0127] When cleaning historical elevator traction machine data, the process begins by filtering out raw data containing key parameters such as vibration, temperature, and rotational speed, and removing records with incorrect data formats. Next, missing values ​​are addressed: if a single parameter is missing from a data entry, it is supplemented with the average of five adjacent data points under the same operating conditions on that day; if more than two parameters are missing, the data entry is deleted. Finally, outliers are identified by calculating the mean and standard deviation of each parameter. Values ​​exceeding the range of "mean ± 3 times standard deviation" are considered outliers and replaced with the mean of the parameter. After cleaning, operating conditions are categorized based on load sensor data into three types: no-load, half-load, and full-load. The cleaned data is then grouped according to operating condition type, and each group is labeled with a corresponding operating condition tag. All labeled data are then integrated to obtain the training sample set for the elevator traction machine.

[0128] When performing multi-scale feature extraction on the training sample set, three time scales and two frequency scales are set. At each time scale, the mean, variance, and peak factor of each parameter in the sample set are calculated; at each frequency scale, the sample data are transformed in the frequency domain to extract the dominant frequency, maximum amplitude, and frequency bandwidth. The features extracted at different scales are classified according to the logic of "time scale-frequency scale-feature type", such as "1 minute-low frequency-vibration mean" and "5 minutes-medium frequency-temperature maximum amplitude". All classified features together constitute a multi-dimensional feature set, resulting in the feature representation space of the elevator traction machine.

[0129] When ranking the importance of features in the feature representation space, the correlation between each feature and elevator traction machine fault records is calculated: the number of faults occurring when the feature value is abnormal is counted, and the number of fault occurrences is divided by the total number of abnormal occurrences for that feature to obtain the feature-fault correlation rate. All features are sorted from high to low according to the feature-fault correlation rate, with the feature with the highest correlation rate at the top, and so on down; if two features have the same correlation rate, they are sorted from smallest to largest according to their corresponding entropy values. The top 80% of the features are extracted and organized into an ordered list to obtain the feature subset of the elevator traction machine.

[0130] When abstracting the feature subset layer by layer, it is carried out in three layers: The first layer is the basic feature abstraction layer, which classifies the single parameter features such as vibration, temperature, and rotational speed in the feature subset and clarifies the normal numerical range of each basic feature; the second layer is the combined feature abstraction layer, which calculates the correlation between basic features, such as the "ratio of peak vibration to mean rotational speed" and the "product of temperature variance and running time", forming combined features and labeling their correspondence with minor faults; the third layer is the health correlation abstraction layer, which matches the combined features with historical fault types to determine the health status corresponding to different combined feature numerical ranges. Through the three layers of abstraction, the correspondence between feature values ​​and health status is established, and the health status mapping relationship of the elevator traction machine is obtained.

[0131] When constructing a hierarchical evaluation network based on the health status mapping relationship, the network consists of four layers: the input layer receives feature data from a subset of features, with each feature corresponding to an input node; the basic evaluation layer connects to the input layer, performs numerical verification on the basic features, determines whether they are within the normal range, and outputs the basic health judgment result; the combined evaluation layer receives the result from the basic evaluation layer, calculates and analyzes the combined features, and outputs the combined health judgment result; the health decision layer integrates the results from the first two layers, combines the health status mapping relationship, and outputs a preliminary health index. By defining the number of nodes in each layer and the connection method between layers, the hierarchical evaluation network for the elevator traction machine is obtained.

[0132] When performing adversarial training on the hierarchical evaluation network, a generator and a discriminator are constructed: the generator generates simulated feature data based on the feature distribution of the training sample set; the discriminator receives real training sample data and the generator's simulated data, and outputs a judgment result of "real" or "simulated". During training, the generator parameters are first fixed, and the discriminator's ability to distinguish between real and simulated data is trained until the discriminator's accuracy reaches 85%; then the discriminator parameters are fixed again, and the generator is trained to generate simulated data that is closer to reality until the discriminator's accuracy drops to around 50%. After 10 rounds of alternating training, training is stopped, and the hierarchical evaluation network at this point is retained, resulting in the initial training model of the hierarchical evaluation network.

[0133] When identifying key feature layers in the initial training model, the basic evaluation layer and the combined evaluation layer are removed sequentially. The error between the model's output health index and the actual health status is then tested. If the error increases by more than 15% after removing the basic evaluation layer, and by more than 20% after removing the combined evaluation layer, the basic evaluation layer and the combined evaluation layer are determined to be key feature layers. When adjusting the network connection structure, the number of neurons in the key feature layers is increased, while the number of connections in non-key layers is reduced. The adjusted model is then retained, resulting in the structurally optimized model of the hierarchical evaluation network.

[0134] Based on the trend of the training loss curve during adversarial training, the parameter update step size of the structure optimization model is adaptively adjusted: if the training loss curve decreases by more than 5% in three consecutive iterations, it indicates that the current step size is too small, and the parameter update step size is increased from 0.01 to 0.015; if the training loss curve increases in two consecutive iterations, it indicates that the current step size is too large, and the step size is decreased from 0.015 to 0.008; if the fluctuation range of the training loss curve is less than 1%, it indicates that the step size is appropriate, and the step size is kept unchanged at 0.008. The step size is dynamically adjusted according to this rule, and training continues for 5 rounds until the training loss value stabilizes below 0.05, at which point training is stopped, and the parameter optimization model of the hierarchical evaluation network is obtained.

[0135] Three expert models were selected, and the core evaluation rules of each model were extracted. For example, in the bearing failure expert model, the rule "vibration frequency between 200-300Hz and amplitude exceeding 0.1mm is considered bearing wear" was defined, and in the motor overheating failure expert model, the rule "temperature exceeding 80℃ for 10 minutes is considered motor overheating" was defined. These rules were transformed into logical code recognizable by the parameter optimization model and embedded into the model's health decision layer. A new "expert rule verification module" was added to perform secondary verification of the health index output by the model. After integration, the health assessment model of the elevator traction machine was obtained.

[0136] The beneficial effects are as follows: data cleaning removes invalid data, and working condition classification ensures sample homogeneity, providing a high-quality data foundation for model training; multi-scale feature extraction and layer-by-layer abstraction comprehensively cover the operating characteristics of the traction machine, ensuring the integrity of feature expression; key layer identification and structural adjustment, and parameter adaptive optimization improve the model's fitting ability; adversarial training enhances the model's generalization ability; and expert model capability transfer supplements professional experience. The final health assessment model can accurately output health indices, effectively improving the accuracy and reliability of elevator traction machine health assessment and providing a scientific basis for equipment maintenance.

[0137] In this embodiment of the invention, S5, the fusion features are input into the health assessment model to generate the health index of the elevator traction machine;

[0138] First, the fused features are preprocessed to ensure they match the format and numerical range of the health assessment model's input layer. For format alignment, each feature data point in the fused features is mapped to a specified node position in the input layer according to the node arrangement order of the health assessment model's input layer. If a fused feature has missing values, it is supplemented with the average value of that feature under the same historical conditions. For numerical normalization, the maximum and minimum values ​​of each fused feature in the historical data are found. The current value of the feature is subtracted from the minimum value, and then divided by the difference between the maximum and minimum values. This maps the values ​​uniformly to the 0-1 range, avoiding the impact of numerical magnitude differences on model calculations. After preprocessing, preprocessed fused features adapted to the model input are obtained.

[0139] The preprocessed fusion features are input into the input layer of the health assessment model. Each feature data corresponds to a node in the input layer, which synchronously transmits the feature data to the basic assessment layer. The basic assessment layer performs numerical verification on each feature data, comparing it to the normal range for that feature. If the value is within the normal range, it outputs a "normal" label and a basic health coefficient of 0.9; if the value exceeds the normal range but does not reach the abnormal threshold, it outputs a "slightly abnormal" label and a basic health coefficient of 0.6; if the value reaches the abnormal threshold, it outputs a "severely abnormal" label and a basic health coefficient of 0.3. The basic health coefficients and labels of all features are summarized and transmitted to the combined assessment layer. The combined assessment layer receives the results from the basic assessment layer and, according to the calculation rules of combined features in the health status mapping relationship, multiplies or adds the basic health coefficients of the associated features to obtain the combined health coefficient. Simultaneously, it corrects the combined health coefficient by considering the fault correlation degree corresponding to the combined features and outputs the corrected combined health coefficient to the health decision layer. The health decision layer collects all combined health coefficients, calculates their average value, and then multiplies it by the model calibration coefficient to obtain the final value. This value is then magnified 100 times to obtain the health index of the elevator traction machine.

[0140] The beneficial effects are as follows: by preprocessing the fusion features through format alignment and numerical normalization, the data quality and adaptability of the input model are ensured, and the evaluation results are avoided from being interfered with by format errors or differences in numerical magnitude. Each layer of the health assessment model processes the data layer by layer according to the established logic, from basic feature verification to combined feature analysis, and then to the final health index calculation. Each step is combined with the health status mapping relationship, so that the generated health index can accurately reflect the actual health status of the elevator traction machine. This provides an objective and reliable quantitative basis for subsequent maintenance assessment based on the health index, and effectively improves the accuracy and practicality of elevator traction machine health assessment.

[0141] S6. Perform a maintenance assessment on the elevator traction machine based on the health index.

[0142] In this embodiment of the invention, the maintenance assessment of the elevator traction machine based on the health index includes:

[0143] Establish a rule base for mapping historical health indices to maintenance strategies;

[0144] Preliminary maintenance recommendations for the elevator traction machine are generated based on the numerical range of the health index.

[0145] Based on the operating environment parameters of the elevator traction machine, the preliminary maintenance recommendations are modified to adapt to the operating conditions, resulting in an environmentally adapted maintenance scheme for the elevator traction machine.

[0146] Based on the aforementioned environment-adaptive maintenance scheme, the execution priority and resource allocation requirements of the maintenance procedures are clearly defined;

[0147] The elevator traction machine is maintained according to the execution priority and resource configuration requirements.

[0148] Collect historical health index data and corresponding maintenance records for the elevator traction machine over the past 5 years. Maintenance records must include maintenance cycles, specific maintenance items, spare parts models used, and equipment runtime after maintenance. Divide the historical health index into four ranges: 80-100 for healthy condition, 60-79 for minor abnormality, 40-59 for moderate abnormality, and 0-39 for severe abnormality. Develop a specific maintenance strategy for each range: For the healthy range, conduct a routine monthly inspection, checking for external damage, abnormal noises during operation, and ensuring the lubricating oil level is within the standard range; for the minor abnormality range, conduct a deep inspection quarterly, adding bearing vibration detection and motor stator temperature measurement to the routine inspection items; for the moderate abnormality range, conduct a deep inspection every two months plus a monthly inspection, replacing aging seals, replenishing special lubricating oil, and calibrating the speed sensor; for the severe abnormality range, immediately shut down for repair, disassembling and inspecting the motor winding insulation, replacing bearings with excessive wear, and readjusting the traction machine's operating parameters. Each interval, corresponding maintenance strategy, maintenance cycle, maintenance items, required spare parts, and maintenance effect are organized into a structured table to form a mapping rule library between the historical health index of the elevator traction machine and the maintenance strategy.

[0149] Obtain the current health index of the elevator traction machine and determine its corresponding interval by referring to the interval division in the mapping rule base. Retrieve the maintenance strategy corresponding to this interval from the rule base, specifying the maintenance cycle, core maintenance items, required spare parts, and expected runtime after maintenance. Organize this information into a standardized text document, clearly annotating the current health index value, the interval it belongs to, the corresponding maintenance cycle, a detailed list of maintenance items, a description of required spare parts, and the expected post-maintenance effects, thus obtaining preliminary maintenance recommendations for the elevator traction machine.

[0150] Collect environmental parameters of the elevator traction machine, including ambient temperature, relative humidity, dust concentration, and average daily operating frequency at the installation location. Develop adaptation rules for environmental parameters and maintenance plans: when the ambient temperature exceeds 25℃, shorten the maintenance cycle by 10%; when the relative humidity exceeds 65%, add an inspection of the integrity of the motor housing's moisture-proof coating; when the dust concentration exceeds 0.3mg / m³, add a dust cleaning item inside the traction machine; when the average daily operating frequency exceeds 120 times, shorten the maintenance cycle by 20%. Revise the initial maintenance recommendations based on the current environmental parameters: the original quarterly maintenance cycle is shortened by 10% due to temperature factors, and further shortened by 20% due to the average daily operating frequency, while adding an inspection of the integrity of the motor housing's moisture-proof coating and a dust cleaning item inside the traction machine. The revised maintenance plan must detail the environmental parameters on which the revision is based, the specific revision content, and the adjusted complete maintenance process, resulting in an environmentally adapted maintenance plan for the elevator traction machine.

[0151] All maintenance procedures are extracted from the environmentally adapted maintenance plan, and their execution priorities are determined according to the fault risk level: bearing vibration value detection has a priority of 1, as abnormal bearings can directly cause traction machine jamming; motor stator temperature measurement has a priority of 2, as excessive temperature can easily burn out the motor; the integrity check of the motor housing moisture-proof coating has a priority of 3, as coating damage can lead to motor moisture absorption; dust cleaning inside the traction machine has a priority of 4, as dust accumulation can accelerate component wear; routine inspection has a priority of 5, with a low risk level. Resources are allocated according to priority: bearing vibration value detection (priority 1) is assigned to 2 technicians with elevator mechanical installation and maintenance certificates, equipped with portable vibration analyzers; motor stator temperature measurement (priority 2) is assigned to 1 technician, equipped with a contact thermometer; motor housing moisture-proof coating inspection (priority 3) is assigned to 1 technician, equipped with a coating thickness gauge; dust cleaning inside the traction machine (priority 4) is assigned to 2 technicians, equipped with low-pressure airflow cleaning guns, non-corrosive cleaning agents, and lint-free cloths; routine inspection (priority 5) is assigned to 1 technician, equipped with a stethoscope and lubricating oil level gauge. Compile a list of process priorities, corresponding technician qualification requirements, number of personnel, and required tool and equipment models and quantities to clarify the resource allocation details for each maintenance process and complete the determination of the execution priority and resource configuration requirements for maintenance processes.

[0152] Develop a maintenance execution plan in descending order of maintenance procedure priority, notify relevant technicians two days in advance to ensure personnel are in place, and perform precision calibration on all tools and equipment. On the day of maintenance, the following steps were performed: Priority 1: bearing vibration value detection. Two technicians attached vibration analyzer probes to the front and rear bearing housings of the traction machine, respectively, and collected three sets of data for each location. The values ​​were recorded and compared with the standard range. If the values ​​were normal, the process proceeded to the next step; if abnormal, the abnormal values ​​were marked and recorded. Next, Priority 2: motor stator temperature measurement. Technicians attached contact thermometer probes to three different locations on the motor stator housing, recorded the values ​​after they stabilized, and compared them with the standard range. Following this, Priority 3: moisture-proof coating inspection. A coating thickness gauge was used to measure the coating thickness in different areas of the motor housing to ensure it met the standard. If the thickness was insufficient, it was marked that recoating was required. Then, Priority 4: internal dust cleaning. Technicians removed the side cover of the traction machine, used a low-pressure airflow cleaning gun to blow away internal dust, and then used a lint-free cloth dampened with cleaning agent to wipe the surfaces of key components such as gears and bearings. Finally, Priority 5: routine inspection. Technicians used a stethoscope to listen to the sound of the traction machine during operation, checked for cracks on the exterior, and measured the lubricating oil level with a level gauge. After each procedure is completed, the technician fills out the maintenance procedure record form, recording the test data, whether there are any abnormalities and the handling status; after all procedures are completed, the technical supervisor reviews the record form, confirms that all maintenance content meets the requirements of the environmental adaptation maintenance plan, and completes the maintenance of the elevator traction machine.

[0153] The beneficial effects include: providing objective data support for maintenance recommendations by establishing a mapping rule base, ensuring that maintenance strategies are based on historical experience; generating preliminary maintenance recommendations according to health index ranges to clarify the direction of basic maintenance, and adjusting the plan according to operating environment parameters to make the plan more in line with the actual working conditions of the equipment, avoiding insufficient or excessive maintenance; clarifying the priority of work processes and resource allocation to ensure that high-risk issues are handled first and resources are allocated reasonably, and finally performing maintenance as required can effectively ensure the healthy operation of elevator traction machines, reduce the failure rate, extend the service life of equipment, reduce maintenance costs, and improve the safety and stability of elevator operation.

[0154] like Figure 2 The diagram shown is a functional block diagram of an elevator traction machine health assessment system based on multi-source data fusion, provided by an embodiment of the present invention.

[0155] The elevator traction machine health assessment system 100 based on multi-source data fusion described in this invention can be installed in an electronic device. Depending on the functions implemented, the elevator traction machine health assessment system 100 may include a multi-source feature extraction module 101, a feature entropy value determination module 102, an entropy weight fusion feature construction module 103, a health assessment model construction module 104, a health index generation module 105, and a maintenance assessment decision module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0156] In this embodiment, the functions of each module / unit are as follows:

[0157] The multi-source feature extraction module 101 is used to extract the time-domain and frequency-domain features of the multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine;

[0158] The feature entropy value determination module 102 is used to determine the entropy value of each feature in the feature set;

[0159] The entropy weight fusion feature construction module 103 is used to assign feature weights to the features based on the information uncertainty of the features represented by the entropy value, and to perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine.

[0160] The health assessment model construction module 104 is used to construct a health assessment model for the elevator traction machine based on historical elevator traction machine data.

[0161] The health index generation module 105 is used to input the fused features into the health assessment model to generate the health index of the elevator traction machine.

[0162] The maintenance assessment decision module 106 is used to conduct a maintenance assessment of the elevator traction machine based on the health index.

[0163] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0164] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0165] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0166] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0167] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for health assessment of elevator traction machines based on multi-source data fusion, characterized in that, The method includes: S1. Extract the time-domain and frequency-domain features of the multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine; S2. Determine the entropy value of each feature in the feature set; S3. Assign feature weights to the features based on the information uncertainty represented by the entropy value, and perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine, including: Based on the information uncertainty of the entropy value representing the feature, the weight coefficients of the feature are assigned to obtain the initial weight allocation scheme of the feature; Perform a weight consistency check on the initial weight allocation scheme to obtain the verified feature weights of the features; The verified feature weights are smoothed to obtain the feature weight distribution of the feature; The features are ranked by importance according to the feature weight distribution to obtain the feature weights of the features; The step of assigning weight coefficients to the features based on the information uncertainty represented by the entropy value, to obtain the initial weight allocation scheme for the features, includes: The entropy value is divided into multiple uncertainty levels and assigned corresponding benchmark weight coefficients; Based on the correlation constraints between features, the baseline weight coefficients are collaboratively optimized to obtain the collaborative weight coefficients of the features. The collaborative weight coefficient is dynamically adjusted based on the accuracy of the historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature; The method of dynamically adjusting the collaborative weight coefficient based on the accuracy of historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature includes: Collect health index output results and actual maintenance records within historical assessment periods to obtain a sample set of assessment accuracy for the aforementioned features; By performing correlation analysis between the evaluation accuracy sample set and the corresponding collaborative weight coefficients, the weight configuration patterns that cause evaluation bias can be identified. Based on the mapping relationship between accuracy and weight adjustment amount, the weight correction parameters of the weight configuration mode are generated; When the accuracy of multiple consecutive evaluation cycles reaches a stable threshold, the weight configuration of the current weight configuration mode is locked to obtain the initial weight allocation scheme of the feature. S4. Based on historical elevator traction machine data, construct a health assessment model for the elevator traction machine; S5. Input the fused features into the health assessment model to generate the health index of the elevator traction machine; S6. Perform a maintenance assessment on the elevator traction machine based on the health index.

2. The elevator traction machine health assessment method based on multi-source data fusion as described in claim 1, characterized in that, The extraction of time-domain and frequency-domain features from multi-source sensor data in the elevator traction machine yields a feature set for the elevator traction machine, including: Signal integrity verification is performed on the multi-source sensor data of the elevator traction machine to obtain the sensor data sequence of the elevator traction machine; Based on the statistical mean, variance, and peak factor of the sensor data sequence, a time-domain feature subset of the elevator traction machine is generated; The frequency domain transformation of the sensor data sequence is performed to obtain the frequency domain feature subset of the elevator traction machine; The time-domain feature subset and the frequency-domain feature subset are combined to generate the feature set of the elevator traction machine.

3. The elevator traction machine health assessment method based on multi-source data fusion as described in claim 1, characterized in that, Determining the entropy value of each feature in the feature set includes: The feature set is segmented to obtain the feature data subsequence of the feature set; The feature data subsequence is divided into states to obtain the state sequence corresponding to the feature data subsequence; The approximate entropy of each feature in the feature set is calculated based on the state transition probability of the state sequence to obtain the entropy value of the elevator traction machine.

4. The elevator traction machine health assessment method based on multi-source data fusion as described in claim 1, characterized in that, The health assessment model for the elevator traction machine, constructed based on historical elevator traction machine data, includes: Data cleaning and operating condition classification were performed on historical elevator traction machine data to obtain the training sample set of the elevator traction machine. Multi-scale feature extraction is performed on the training sample set to obtain the feature representation space of the elevator traction machine; The features in the feature representation space are ranked by importance to obtain the feature subset of the elevator traction machine; By performing layer-by-layer feature abstraction on the feature subset, the health status mapping relationship of the elevator traction machine is obtained; A hierarchical evaluation network for the elevator traction machine is constructed based on the health status mapping relationship; The hierarchical evaluation network is trained and optimized to form the health assessment model for the elevator traction machine.

5. The elevator traction machine health assessment method based on multi-source data fusion as described in claim 4, characterized in that, The training and optimization of the hierarchical evaluation network, and the resulting health assessment model for the elevator traction machine, include: Adversarial training is performed on the hierarchical evaluation network to obtain the initial training model of the hierarchical evaluation network; Identify the key feature layers in the initial training model and adjust the network connection structure of the initial training model to obtain the structure optimization model of the hierarchical evaluation network; Based on the changing trend of the training loss curve of the adversarial training, the parameter update step size of the structure optimization model is adaptively adjusted to obtain the parameter optimization model of the hierarchical evaluation network. The evaluation capabilities of multiple expert models are transferred to the parameter optimization model to obtain the health assessment model of the elevator traction machine.

6. The elevator traction machine health assessment method based on multi-source data fusion as described in claim 1, characterized in that, The maintenance assessment of the elevator traction machine based on the health index includes: Establish a rule base for mapping historical health indices to maintenance strategies; Preliminary maintenance recommendations for the elevator traction machine are generated based on the numerical range of the health index. Based on the operating environment parameters of the elevator traction machine, the preliminary maintenance recommendations are modified to adapt to the operating conditions, resulting in an environmentally adapted maintenance scheme for the elevator traction machine. Based on the aforementioned environment-adaptive maintenance scheme, the execution priority and resource allocation requirements of the maintenance procedures are clearly defined; The elevator traction machine is maintained according to the execution priority and resource configuration requirements.

7. A health assessment system for elevator traction machines based on multi-source data fusion, used to implement the health assessment method for elevator traction machines based on multi-source data fusion as described in claim 1, the system comprising: A multi-source feature extraction module is used to extract the time-domain and frequency-domain features of multi-source sensor data in the elevator traction machine to obtain the feature set of the elevator traction machine; The feature entropy value determination module is used to determine the entropy value of each feature in the feature set; The entropy-weighted fusion feature construction module is used to assign feature weights to the features based on the information uncertainty represented by the entropy value, and to perform weighted fusion of the features based on the feature weights to obtain the fused features of the elevator traction machine. Specifically, it is used for: Based on the information uncertainty of the entropy value representing the feature, the weight coefficients of the feature are assigned to obtain the initial weight allocation scheme of the feature; Perform a weight consistency check on the initial weight allocation scheme to obtain the verified feature weights of the features; The verified feature weights are smoothed to obtain the feature weight distribution of the feature; The features are ranked by importance according to the feature weight distribution to obtain the feature weights of the features; The step of assigning weight coefficients to the features based on the information uncertainty represented by the entropy value, to obtain the initial weight allocation scheme for the features, includes: The entropy value is divided into multiple uncertainty levels and assigned corresponding benchmark weight coefficients; Based on the correlation constraints between features, the baseline weight coefficients are collaboratively optimized to obtain the collaborative weight coefficients of the features. The collaborative weight coefficient is dynamically adjusted based on the accuracy of the historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature; The method of dynamically adjusting the collaborative weight coefficient based on the accuracy of historical evaluation results in the elevator traction machine to obtain the initial weight allocation scheme for the feature includes: Collect health index output results and actual maintenance records within historical assessment periods to obtain a sample set of assessment accuracy for the aforementioned features; By performing correlation analysis between the evaluation accuracy sample set and the corresponding collaborative weight coefficients, the weight configuration patterns that cause evaluation bias can be identified. Based on the mapping relationship between accuracy and weight adjustment amount, the weight correction parameters of the weight configuration mode are generated; When the accuracy of multiple consecutive evaluation cycles reaches a stable threshold, the weight configuration of the current weight configuration mode is locked to obtain the initial weight allocation scheme of the feature. The health assessment model construction module is used to construct a health assessment model for the elevator traction machine based on historical elevator traction machine data. A health index generation module is used to input the fused features into the health assessment model to generate the health index of the elevator traction machine; The maintenance assessment and decision-making module is used to assess the maintenance of the elevator traction machine based on the health index.

Images