Elevator door motor system fault diagnosis method and system based on deep learning

Abstract

The invention discloses an elevator door motor system fault diagnosis method and system based on deep learning, and relates to the technical field of elevator door motor system fault diagnosis. According to the elevator door motor system fault diagnosis method based on deep learning, operation state time sequence data of all operation stages are collected, a pre-training model is input to extract features, an operation health time sequence index is constructed, whether a fault exists or not is judged in combination with a preset health interval, and multi-stage and intelligent door motor fault recognition is achieved; according to the method, the operation state time sequence data of the elevator door motor in each operation stage are obtained, the time sequence modeling capacity of the deep learning model is combined, dynamic modeling of the operation state of the door motor system in the whole period is achieved, abnormal segments are highlighted through a multi-head attention mechanism, key operation features are extracted, and the operation state of the elevator door motor system is obtained. And the staged health index of each operation stage is output, and the operation health time sequence index of each stage is compared with a preset health interval for judgment, so that finer-grained stage-by-stage fault identification can be realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of elevator door machine system fault diagnosis, and particularly relates to an elevator door machine system fault diagnosis method and system based on deep learning. BACKGROUND

[0002] As a key executive component in the safety of elevator operation, the elevator door machine system is responsible for completing the opening and closing operation of the car door and the landing door, and its working state directly affects the passenger boarding safety and equipment operation stability. The existing elevator door machine adopts a motor driving structure, and the opening and closing speed, position and response rhythm are adjusted through a control module. In actual operation, the elevator door machine is easily affected by factors such as mechanical wear, track deformation, electrical fluctuation and control delay, resulting in door jamming, deviation, misoperation and other fault phenomena.

[0003] The prior art such as the elevator door system fault diagnosis method and system based on sound recognition technology disclosed in the patent application with the announcement number CN116101864B comprises the following steps: S1, when the car door of the elevator car is opened and closed, sound data, acceleration data of the car door and vibration data of the elevator door machine of the elevator door system are collected respectively, and the collected sound data, acceleration data and vibration data are sent to an industrial computer; S2, the industrial computer runs a sound algorithm model and periodically analyzes the sound data of the elevator door system. According to the application, the sound, acceleration and vibration data of the elevator door system are collected by various sensors, and sound recognition and fault judgment algorithms are used to analyze and judge whether the elevator door system is abnormal, so that the health state of the elevator door system can be quickly and accurately detected and diagnosed, and the health state can be displayed on a cloud platform. Remote monitoring of the elevator door system can be realized, and the elevator door system can automatically alarm when it is abnormal.

[0004] Based on the above scheme discovery, the limitations of the prior art include at least the following problems. In the prior art, the elevator door system fault diagnosis method based on sound recognition has significant structural technical limitations. The root cause lies in that the fault judgment process highly depends on the mutation of acoustic signal features and the pattern classification of single-time local features. First, this kind of method usually samples sound from a single channel, constructs a feature vector by analyzing the sound signal amplitude, spectral features or kurtosis at a specific time point, and uses it to determine whether an abnormality occurs. However, there are multiple sound interferences in the actual operation of the elevator door machine, such as background noise, control action superposition, structural reflection echo, and the sound characteristics of different types of door machines in normal state are significantly different, which leads to that this kind of method is highly sensitive to the field environment and equipment type, has poor stability and weak generalization ability. Second, the sound recognition method only relies on short-time transient signals in the door opening and closing process, and it is difficult to systematically capture the state evolution of the door machine in multiple stages such as starting, accelerating, uniform speed, decelerating and stopping. Its judgment basis is often concentrated on whether an abnormal sound occurs, which is a point event. It lacks continuous monitoring and reasoning ability for physical variables such as motor current, voltage change, control response delay, door body friction trend and structural deviation, which makes it difficult to identify and trend early warning for structural faults, control chain delay drift or gradual wear. Third, the threshold judgment and classification rules used in the existing method are usually based on experience setting or specific sample training, and cannot build a unified state health expression model. It is also difficult to infer the overall running state of the door machine system through the time sequence correlation of cross-stage features, which limits the diagnosis range to the recognition of sudden abnormalities, and lacks effective capture ability for long-period system performance degradation, behavior deviation and other non-explicit faults. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a deep learning-based elevator door machine system fault diagnosis method and system, which solves the problem of the prior art that only relies on local sound mutation for static classification and lacks modeling of multiple physical features throughout the door machine operation process and time sequence health trend judgment capability.

[0006] To achieve the above object, the application is implemented by the following technical solutions: A deep learning-based elevator door machine system fault diagnosis method, comprising the following steps: continuously acquiring running state time series data of each running stage of the elevator door machine; inputting the running state time series data of each running stage of the elevator door machine into a pre-trained deep learning model respectively for feature extraction processing to obtain running time series features of each running stage of the elevator door machine; analyzing running health time series indexes of each running stage of the elevator door machine based on the running time series features of each running stage of the elevator door machine; judging and analyzing the running health time series indexes of each running stage of the elevator door machine with corresponding preset stage running health intervals respectively, and regarding the running stage with the running health time series index outside the preset stage running health interval as a running fault.

[0007] Further, the running state time series data includes motor running current, motor running voltage, motor running power, door machine electromagnetic interference intensity, control response delay, door body friction load, and door body offset angle at several time points.

[0008] Further, the deep learning model includes an input embedding layer, a bidirectional GRU layer, a multi-head attention layer, a stage feature aggregation layer, and a feature output layer, and the running time series features include motor running current peak value, motor running current fluctuation variance, motor running current high-frequency disturbance times, motor running voltage stability coefficient, door machine electromagnetic interference average intensity, door machine electromagnetic interference duration, control response average delay, maximum control response delay, control response delay fluctuation rate, door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency.

[0009] Further, the specific steps of obtaining the running time sequence feature of each running stage of the elevator door machine are as follows: in the input embedding layer of the deep learning model, the running state time sequence data of each running stage of the elevator door machine is received, and normalization processing and feature dimension mapping processing are performed to obtain the running time sequence vector sequence of each running stage of the elevator door machine with uniform dimensions; in the bidirectional GRU layer of the deep learning model, the running time sequence vector sequence of each running stage of the elevator door machine is subjected to time sequence modeling processing to capture the forward and backward dependence relationship of the running state, extract the context state code at each time point, and obtain the time sequence hidden state vector set of each running stage of the elevator door machine; in the multi-head attention layer of the deep learning model, the time sequence hidden state vector set of each running stage of the elevator door machine is subjected to parameter importance evaluation and time slice abnormal segment focusing processing to extract the cross feature representation of the high weight running segment, and obtain the stage running state feature vector set of each running stage of the elevator door machine with attention weighting mechanism; in the stage feature gathering layer of the deep learning model, the stage running state feature vector set of each running stage of the elevator door machine after attention weighting is subjected to global average gathering processing to form the stage comprehensive representation vector corresponding to the running stage; in the feature output layer of the deep learning model, the stage comprehensive representation vector of each running stage of the elevator door machine is subjected to mapping processing to output the running time sequence feature set of each running stage of the elevator door machine.

[0010] Further, the specific steps of analyzing the running health time sequence index of each running stage of the elevator door machine are as follows: based on the running time sequence feature of each running stage of the elevator door machine, the driving system health index, the control response stability index, and the door body structure posture index of each running stage of the elevator door machine are analyzed respectively; the driving system health index, the control response stability index, and the door body structure posture index of each running stage of the elevator door machine are comprehensively analyzed respectively to obtain the running health time sequence index of each running stage of the elevator door machine.

[0011] Further, the specific formula for calculating the running health time sequence index of a certain running stage of the elevator door machine is as follows: ; wherein, is the running health time sequence index of a certain running stage of the elevator door machine, is the driving system health index of a certain running stage of the elevator door machine, is a natural constant, is the door body structure posture index of a certain running stage of the elevator door machine, is the structure health adjustment coefficient stored in the database, is the control response stability index of a certain running stage of the elevator door machine, is the control coordination enhancement coefficient stored in the database.

[0012] Further, the specific steps of analyzing the drive system health index of each running stage of the elevator door machine are as follows: reading the motor running current peak value, motor running current fluctuation variance, motor running current high frequency disturbance times, and motor running voltage stability coefficient of each running stage of the elevator door machine; comprehensively analyzing the motor running current peak value, motor running current fluctuation variance, motor running current high frequency disturbance times, and motor running voltage stability coefficient of each running stage of the elevator door machine respectively to obtain the drive system health index of each running stage of the elevator door machine.

[0013] Further, the specific steps of analyzing the control response stability index of each running stage of the elevator door machine are as follows: reading the door machine electromagnetic interference average intensity, door machine electromagnetic interference duration, control response average delay, maximum control response delay, and control response delay fluctuation rate of each running stage of the elevator door machine; comprehensively analyzing the door machine electromagnetic interference average intensity, door machine electromagnetic interference duration, control response average delay, maximum control response delay, and control response delay fluctuation rate of each running stage of the elevator door machine respectively to obtain the control response stability index of each running stage of the elevator door machine.

[0014] Further, the specific steps of analyzing the door body structure posture index of each running stage of the elevator door machine are as follows: reading the door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency of each running stage of the elevator door machine; comprehensively analyzing the door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency of each running stage of the elevator door machine respectively to obtain the door body structure posture index of each running stage of the elevator door machine.

[0015] A deep learning-based elevator door machine system fault diagnosis system, comprising: a data acquisition unit configured to continuously acquire running state time series data of each running stage of the elevator door machine; a feature extraction unit configured to input the running state time series data of each running stage of the elevator door machine into a pre-trained deep learning model for feature extraction processing to obtain running time series features of each running stage of the elevator door machine; a running health analysis unit configured to analyze running health time series indexes of each running stage of the elevator door machine based on the running time series features of each running stage of the elevator door machine; and a running fault analysis unit configured to judge and analyze the running health time series indexes of each running stage of the elevator door machine with respect to corresponding preset stage running health intervals, and regard the running stage outside the preset stage running health interval as a running fault.

[0016] The present application has the following beneficial effects: (1) The elevator door machine system fault diagnosis method based on deep learning, by continuously acquiring the running state time series data of the elevator door machine in each running stage, and combining the time series modeling ability of the deep learning model, the dynamic modeling of the running state of the door machine system in the whole cycle is realized for the first time, compared with the existing method which depends on the static sound mutation or rule threshold judgment, this scheme uses the pre-trained deep learning model to forward and backward model the time series composed of multi-dimensional physical quantities, further highlights the abnormal fragments and extracts the key running features through the multi-head attention mechanism, finally outputs the stage health index of each running stage, by comparing the running health time series index of each stage with the preset health interval, more fine-grained stage-by-stage fault identification can be realized, avoiding the diagnosis deviation caused by ignoring the stage difference in the existing method, significantly improving the recognition accuracy and response timeliness of the slowly changing and non-sudden door machine faults.

[0017] (2) The elevator door machine system fault diagnosis method based on deep learning, by taking the running state time series data of the running stage as the time series input, it no longer depends on non-steady signals such as sound and vibration which are easily affected by the environment, after the deep learning feature extraction is completed, the driving system health index, control response stability index and door body structure posture index are constructed to modularize and model various core influencing factors, and a unified running health time series index is formed, and the adaptability of the evaluation result to different door machine types and use scenarios is improved through the adjustment coefficient obtained by training in the database, solving the problem of poor model migration and weak generalization ability of the existing scheme, and having the ability of cross-scene deployment.

[0018] (3) The elevator door machine system fault diagnosis method based on deep learning, by introducing a deep learning model with clear structure, including input embedding layer, bidirectional GRU layer, multi-head attention layer, stage feature aggregation layer and feature output layer, this model not only can model the high-dimensional vectorization of the normalized multi-dimensional original running state, but also can strengthen the recognition ability of short-time mutation and chronic risk trend through attention mechanism, finally output the semantic fault feature set of the elevator door machine running stage, on this basis, the structure adjustment coefficient and control enhancement coefficient are trained through the preset objective function and historical health level data, so that the system has the ability to dynamically adjust the health index weight in different use environments, compared with the existing static rule or experience judgment scheme, it has significant self-adaptive learning ability and intelligent fault identification ability, can realize online fault level evaluation and early warning strategy linkage deployment, and improve the elevator operation safety protection ability.

[0019] (4), the elevator door machine fault diagnosis system based on deep learning, through structuring the elevator door machine fault diagnosis process and dividing it into four functional modules of data acquisition unit, feature extraction unit, running health analysis unit and running fault analysis unit, a modular intelligent diagnosis system architecture with clear boundaries and data interface is constructed, the modules are connected through standardized running stage index and feature data vector, the system has good functional independence and deployment flexibility, the data acquisition unit can connect multiple brand door machine systems and collect uniform standard physical parameter time series data, the feature extraction unit realizes seamless embedded feature extraction based on pre-trained model, the running health analysis and fault analysis unit realizes logical judgment based on standardized index and preset interval, the system design simplifies subsequent model replacement, function upgrade and multi-scene adaptation work, solves the problems of high functional coupling, difficult module replacement and operation and maintenance separation of existing fault detection systems, and has good engineering implementation and maintenance scalability.

[0020] Of course, implementing any product of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flow chart of a method for diagnosing faults of an elevator door machine system based on deep learning.

[0022] Figure 2 A flow chart of specific steps for analyzing running health time series indexes of each running stage of an elevator door machine in a method for diagnosing faults of an elevator door machine system based on deep learning.

[0023] Figure 3 A block diagram of a system for diagnosing faults of an elevator door machine system based on deep learning. DETAILED DESCRIPTION

[0024] Please refer to Figure 1 The embodiment of the present application provides a technical scheme: a method for diagnosing faults of an elevator door machine system based on deep learning, comprising the following steps: continuously acquiring running state time series data of each running stage (for example: starting, accelerating, uniform speed, decelerating, stopping) of an elevator door machine; inputting the running state time series data of each running stage of the elevator door machine into a pre-trained deep learning model respectively for feature extraction processing, to obtain running time series features of each running stage of the elevator door machine; analyzing running health time series indexes of each running stage of the elevator door machine based on the running time series features of each running stage of the elevator door machine; judging and analyzing the running health time series indexes of each running stage of the elevator door machine respectively with respect to corresponding preset stage running health intervals, and regarding the running stage with a running health time series index outside the preset stage running health interval as a running fault.

[0025] The running state timing data includes motor running current, motor running voltage, motor running power, door machine electromagnetic interference intensity, control response delay, door body friction load, and door body offset angle at several time points.

[0026] The motor running current represents the instantaneous current value of the elevator door machine motor during the running process, reflects the energy supply intensity of the drive system to the load, and indicates system overload, abnormal start or electrical control instability when the current is too large or fluctuates sharply. The acquisition step is as follows: during data collection, the current signal of each running stage is continuously sampled by the current collection module installed at the input end of the door machine motor, the sampling period is set according to the control system synchronization signal (for example, every 0.1 second), the collected data is transmitted to the local controller in real time and stored in the database, forming a complete current timing sequence for subsequent normalization and feature extraction processing.

[0027] The motor running voltage represents the actual power supply voltage of the elevator door machine motor during the running process, reflects the stability of the power supply system, and may cause control error, power deviation or electrical fault when the fluctuation is large. The acquisition step is as follows: the voltage acquisition module synchronized with the current is used to sample the high frequency of the door machine motor power supply loop, record the voltage value at each time point, and transmit it to the system control module after analog-digital conversion. All sampling points are marked with running stage time stamp and stored in the database for evaluating voltage stability and calculating electric power.

[0028] The motor running power represents the instantaneous output power of the motor at a certain time point, is used to measure the load matching state and energy consumption of the motor, and is one of the basic parameters for judging whether the system is overloaded or inefficiently running. The acquisition step is as follows: the motor running power is obtained by multiplying the collected motor running current I(t) and running voltage V(t) point by point; the calculation is completed in real time at the system control end, and the result is bound with the original sampling time and written into the database to form a complete power timing curve.

[0029] The door machine electromagnetic interference intensity represents the detected electromagnetic interference amplitude during the running process of the door machine, which may be caused by switch action, motor commutation or external device interference. Over strong interference may affect the transmission of control command and the reliability of sensor collection. The acquisition step is as follows: the electromagnetic interference detection unit (such as near field EMI sensor or shielding loop detector) installed on the door machine control board or its periphery is used to collect the interference signal intensity in real time, and the sampling frequency is set to be synchronized with the control main loop; each sampling result is stored as a timing sequence for subsequent calculation after filtering and amplitude conversion.

[0030] The control response delay represents the time difference between the time when the system issues the door control signal and the time when the door body actually starts to execute the action, which is a key parameter for measuring the execution efficiency and response consistency of the control link. The acquisition step is as follows: the control response delay is obtained by recording the time stamp Tcmd corresponding door body action detection feedback timestamp T resp , the delay value Δ = T resp -T cmd is calculated. The whole process is completed through controller logging, encoder signal comparison, or door body start sensor determination, and is stored separately according to the running stage.

[0031] Door body friction load, which represents the resistance level of the door body during movement, is a comprehensive reflection of structural wear, track impurities, and lubrication conditions; excessive friction will lead to slow operation, abnormal heating, or jamming; the acquisition steps are as follows: the friction resistance is inferred from the instantaneous current, voltage, and actual motion state of the door machine driving motor (such as based on motor model or known power-speed relationship), and the running state is determined by combining the door body position encoder or acceleration sensor; the system continuously stores the obtained friction value according to time, and uses it for slope and abnormal change analysis.

[0032] Door body offset angle, which represents the deviation angle of the door body from the ideal trajectory during track operation, continuous deviation may indicate track bending, wheel system loosening, or structural misalignment; the acquisition steps are as follows: the angle change data of the door body during operation is obtained through the inertial measurement unit (IMU) or attitude sensor installed on the left and right guide rails or the frame structure of the door body; each record includes pitch, roll, or yaw angle; the system normalizes the offset angle data and writes it into the database to form a set of attitude angle time series.

[0033] Specifically, the deep learning model includes an input embedding layer, a bidirectional GRU layer, a multi-head attention layer, a stage feature aggregation layer, and a feature output layer; the running time series features include motor running current peak value, motor running current fluctuation variance, motor running current high-frequency disturbance frequency, motor running voltage stability coefficient, door machine electromagnetic interference average intensity, door machine electromagnetic interference duration, control response average delay, maximum control response delay, control response delay fluctuation rate, door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency.

[0034] The pre-training steps of the deep learning model are as follows: A large number of elevator door machine running stage data covering normal operation state and labeled fault samples are collected, which include motor current, voltage, power, control delay, EMI intensity, door body friction load, and offset angle at multiple time points, and are archived according to running stage labels (such as start, acceleration, deceleration, and stop) and health state labels (such as normal / abnormal / mild abnormal).

[0035] The above data is normalized, the dimension and scale are unified, and the training set, validation set and test set are divided. Each time series data is labeled according to the running stage, so that the model can perceive the stage information and the corresponding relationship between the target output during training.

[0036] The model training target is set as a multi-task loss: The running feature reconstruction loss is measured by using mean square error or Huber loss to measure the difference between the physical quantity feature vector output by the model and the real label.

[0037] The normalized time series data is input into the model structure, and is sequentially processed by input embedding, bidirectional GRU modeling, multi-head attention extraction, stage feature aggregation and feature decoding output. The Adam or RMSprop optimizer is used to perform gradient descent training on the model parameters until the loss converges.

[0038] After training is completed, the model structure and parameter weights are exported and deployed in the system as a pre-trained model. Subsequently, in the real running environment, no further training is required, and only the feature extractor is used. The model can also be retrained or fine-tuned at a set period to improve adaptability.

[0039] The motor running current peak value represents the maximum value of the current during the running stage, reflecting the instantaneous strength of the start or load.

[0040] The motor running current fluctuation variance represents the fluctuation degree of the current during the running process. The larger the value, the worse the power supply stability.

[0041] The motor running current high-frequency disturbance frequency represents the number of times the current jumps frequently in a short time, reflecting the power supply interference or control signal jitter.

[0042] The motor running voltage stability coefficient represents the stability of the voltage around the rated value. The closer to 1, the better, and the smaller the fluctuation, the more stable the system.

[0043] The door machine electromagnetic interference average intensity represents the average amplitude of the detected electromagnetic field interference during the running period, reflecting the electromagnetic pollution level.

[0044] The door machine electromagnetic interference duration represents the total time of EMI interference affecting system control or electrical stability.

[0045] The control response average delay represents the average time from the issuance of the control command to the response of the door machine action. The lower the value, the more agile.

[0046] The maximum control response delay represents the maximum response delay encountered during the entire running stage, reflecting the extreme instability.

[0047] Control response delay fluctuation rate, representing the fluctuation degree of control delay, reflecting response consistency and control chain stability.

[0048] Door body static friction load, representing the resistance load in the stationary state before door body start, determining whether the start is laborious or stuck.

[0049] Door body dynamic friction load, representing the friction resistance encountered by the door body in operation, reflecting track wear or foreign matter problems.

[0050] Door body friction trend slope, representing the trend of friction force change over time, indicating structure wear or lubrication deterioration.

[0051] Door body offset mutation rate, representing the frequency of short-term sharp mutation of door body posture angle, representing structure loosening or imbalance impact.

[0052] Door body offset average angle, representing the average offset angle of the door body during the entire operation process.

[0053] Door body jitter frequency, representing the frequency of micro-vibration or mechanical structure vibration of the door body in operation, the higher the frequency, the more unstable the structure.

[0054] The specific steps of obtaining the running time sequence characteristics of each running stage of the elevator door machine are as follows: in the input embedding layer of the deep learning model, the running state time sequence data of each running stage of the elevator door machine is received, and normalization processing and feature dimension mapping processing are performed, to obtain the running time sequence vector sequence of the elevator door machine in each running stage with uniform dimensions, which is specific: receiving the original running state time sequence data of the elevator door machine in each running stage, the time sequence data including the motor running current, motor running voltage, motor running power, door machine electromagnetic interference intensity, control response delay, door body friction load, and door body offset angle of several time points. First, the above original physical data is normalized to eliminate the influence caused by the differences in dimensions, units or values between different physical quantities, so as to adapt to a unified calculation space. Then, a set of linear mapping layers or one-dimensional convolution kernels are used to perform feature dimension mapping processing on the parameter vector of each time point, to expand the original 7-dimensional physical parameters to a unified high-dimensional embedding vector (such as 64-dimensional), and to form a running time sequence vector sequence with a unified structure, providing a semantically consistent representation basis for subsequent time sequence modeling; in the bidirectional GRU layer of the deep learning model, the running time sequence vector sequence of each running stage of the elevator door machine is processed by time sequence modeling to capture the forward and backward dependence relationship of the running state, and to extract the context state encoding of each time point, to obtain the time sequence hidden state vector set of each running stage of the elevator door machine, which is specific: receiving the embedded running time sequence vector sequence, and based on the gated recurrent neural network structure, processing the bidirectional time sequence modeling, on one hand, the forward GRU unit is used to model the time sequence from front to back, to capture the forward state evolution trend of each time point in the running process of the elevator door machine; on the other hand, the reverse GRU unit is used to model the time sequence from back to front, to enhance the model's ability to perceive the influence of subsequent changes on the previous state, and finally, the forward and reverse hidden states are spliced and fused to form the time sequence hidden state vector set of the current running stage of the elevator door machine, so that the model can understand the dynamic evolution relationship of the context state at each time step.In the multi-head attention layer of the deep learning model, the parameter importance evaluation and time slice abnormal segment focusing processing are performed on the time sequence hidden state vector set of each operating phase of the elevator door machine, the cross feature representation of the high weight operating segment is extracted, and the phase operating state feature vector set of each operating phase of the elevator door machine with attention weighting mechanism is obtained, which is specifically: receiving the time sequence hidden state vector set, and using the multi-head attention mechanism to perform time sequence information importance evaluation and cross feature modeling processing, specifically, the layer maps the time sequence hidden state sequence into query vector, key vector and value vector respectively, calculates the dependence weight between each time point through the scaling dot product attention mechanism, and learns the importance distribution under different perspectives based on multiple subspaces (i.e. multi-head structure), thereby improving the focusing ability of the model in identifying key segments such as high-risk segments, short-time mutation segments or long-term drift trends in operation, and finally, fusing the multi-head attention output to form a phase operating state feature vector sequence containing attention weighting information, providing a structure-enhanced representation basis for subsequent global aggregation and feature decoding; in the phase feature aggregation layer of the deep learning model, the global average aggregation processing is performed on the phase operating state feature vector set of each operating phase of the elevator door machine after attention weighting, and the phase comprehensive representation vector corresponding to the operating phase is formed, which is specifically: performing global average aggregation processing on the phase operating state feature vector sequence after multi-head attention processing, extracting the feature expression of the operating phase in the overall time span, specifically, using the global average pooling (Global Average Pooling) method to compress the time dimension, performing dimension-by-dimension averaging operation on the vectors of all time points, thereby obtaining a fixed-dimension phase comprehensive representation vector, which effectively compresses the dynamic evolution information, important state response and parameter fluctuation mode of the current operating phase, has global expression ability, and can be used as a semantic carrier for fault mode analysis, health score and feature decoding; in the feature output layer of the deep learning model, the mapping processing is performed on the phase comprehensive representation vector of each operating phase of the elevator door machine, and the operating time sequence feature set of each operating phase of the elevator door machine is output, which is specifically: inputting the phase comprehensive representation vector into the multi-channel mapping structure for explicit decoding processing of multi-physical quantity features, the output layer expands the phase vector through the fully connected network or feature branch module, and outputs a set of physical feature values highly related to the actual operating state, the operating time sequence feature set includes but is not limited to: motor operating current peak value, current fluctuation variance, current high-frequency disturbance frequency, voltage stability coefficient, electromagnetic interference average intensity and duration, control response average delay, maximum delay and delay fluctuation rate, door body static friction and dynamic friction load, friction trend slope, door body offset angle mean, offset mutation rate and door body jitter frequency, etc. The feature set as a high semantic output result can be directly used for subsequent operating health index modeling and fault level judgment processing.

[0055] In the embodiment, the proposed deep learning model structure is clear and hierarchical, which can automatically extract high semantic operation features from the original physical time series data. Through input embedding, bidirectional GRU modeling, multi-head attention focusing, stage feature aggregation and structured decoding, the model realizes comprehensive understanding and fine-grained description of the state of each operation stage of the elevator door machine. The method does not rely on artificial rules, has good context modeling ability and key moment recognition ability, and the extracted operation time series features have high interpretability and fault correlation, which provides a reliable data basis for subsequent health index construction and fault level judgment, and significantly improves the intelligence and adaptability of the system.

[0056] Specifically, as shown in Figure 2 , the specific steps of analyzing the operation health time series index of each operation stage of the elevator door machine are as follows: based on the operation time series features of each operation stage of the elevator door machine, the drive system health index, the control response stability index, and the door body structure posture index of each operation stage of the elevator door machine are analyzed respectively; the drive system health index, the control response stability index, and the door body structure posture index of each operation stage of the elevator door machine are analyzed respectively, and the operation health time series index of each operation stage of the elevator door machine is obtained.

[0057] The specific formula for calculating the operation health time series index of a certain operation stage of the elevator door machine is as follows: ; wherein, is the operation health time series index of a certain operation stage of the elevator door machine, is the drive system health index of a certain operation stage of the elevator door machine, is a natural constant, which is 2.71 in this embodiment, is the door body structure posture index of a certain operation stage of the elevator door machine, is the structure health adjustment coefficient stored in the database, is the control response stability index of a certain operation stage of the elevator door machine, is the control coordination enhancement coefficient stored in the database.

[0058] It should be explained that the structure health adjustment coefficient stored in the database, the control coordination enhancement coefficientThe specific acquisition steps are: first, based on the elevator door machine fault records of multiple running stages in the historical samples and the time series data of the corresponding drive system health index, control response stability index and door body structure posture index, a set of standard running data set for parameter training is constructed, secondly, the data set is input into the pre-defined target function, the target function is used to minimize the deviation between the running health time series index and the actual artificial judgment health level, on this basis, the gradient descent type optimization algorithm (such as Adam optimizer or adaptive momentum method) is used to jointly update the structure health adjustment coefficient and control coordination enhancement coefficient, and finally the coefficient value that makes the running health time series index and the actual health state most consistent is converged, the parameter acquisition process can be completed once in the system initialization stage, or it can be dynamically adjusted periodically combined with real-time running data, to ensure that the running health time series index evaluation mechanism has adaptability and accuracy under different elevator models or running conditions.

[0059] Among them, the specific implementation example of calculating the running health time series index of a running stage of the elevator door machine is as follows, the existing parameters are as follows: The drive system health index of a running stage of the elevator door machine is about 0.742.

[0060] Natural constant The value is: 2.71.

[0061] The door body structure posture index of a running stage of the elevator door machine is about 0.435.

[0062] The structure health adjustment coefficient stored in the database is about 1.215.

[0063] The control response stability index of a running stage of the elevator door machine is about 0.668.

[0064] The control coordination enhancement coefficient stored in the database is about 1.034.

[0065] The above data is substituted into the specific formula for calculating the running health time series index of a running stage of the elevator door machine respectively, and the following is obtained: The running health time series index of a running stage of the elevator door machine = (( (0.742 x 2.71^ (-0.435))^1.215) + (( (0.668^2) / (1 + |0.742-0.668|))^1.034))^0.5≈0.902.

[0066] In this embodiment, by introducing a three-dimensional decomposition mechanism for driving system health index, control response stability index and door body structure posture index, the structural and multi-angle evaluation of the elevator door machine running state is realized. At the same time, the structural health adjustment coefficient and control coordination enhancement coefficient trained based on historical sample data are used to make the calculation of the comprehensive running health time sequence index have data-driven adaptive ability. This method not only enhances the consistency between the index result and the actual fault level, but also can be dynamically adjusted according to different elevator models and running conditions, significantly improving the accuracy, interpretability and cross-scene applicability of the evaluation index, and providing a stable, adjustable and transferable core support mechanism for system intelligent diagnosis.

[0067] Specifically, the specific steps of analyzing the driving system health index of each running stage of the elevator door machine are as follows: reading the motor running current peak value, motor running current fluctuation variance, motor running current high frequency disturbance times and motor running voltage stability coefficient of each running stage of the elevator door machine; comprehensively analyzing the motor running current peak value, motor running current fluctuation variance, motor running current high frequency disturbance times and motor running voltage stability coefficient of each running stage of the elevator door machine, respectively, to obtain the driving system health index of each running stage of the elevator door machine.

[0068] Wherein, the specific formula for calculating the driving system health index of a certain running stage of the elevator door machine is as follows: ; wherein, is the driving system health index of a certain running stage of the elevator door machine, is a natural constant, is the motor running current peak value of a certain running stage of the elevator door machine, is the current peak value adjustment coefficient stored in the database, is the motor running current fluctuation variance of a certain running stage of the elevator door machine, is the running current adjustment factor stored in the database, which is used to prevent the denominator from being 0, is the motor running voltage stability coefficient of a certain running stage of the elevator door machine, is the voltage stability adjustment coefficient stored in the database, is the motor running current high frequency disturbance times of a certain running stage of the elevator door machine.

[0069] It needs to be explained that the current peak value adjustment coefficient stored in the database The specific acquisition steps are: first, extracting the complete time sequence of the motor operating current and voltage corresponding to each operating stage in the historical data; second, normalizing the current peak value and voltage stability in each period respectively, calculating the relative fluctuation degree of the current peak value and the amplitude of the voltage value deviating from the rated value; and then performing mean value processing on multiple operating samples to construct the current peak value adjustment coefficient and the voltage stability adjustment coefficient in the database, respectively, as adjustment parameters for the influence degree of current and voltage on the driving system health status.

[0070] In this embodiment, by introducing the driving system health index constructed based on physical characteristic parameters, the motor operating current peak value, current fluctuation variance, current high-frequency disturbance frequency and voltage stability and other key indicators are systematically integrated, comprehensively reflecting the load pressure and power supply stability of the driving system in each operating stage. At the same time, with the aid of the current peak value adjustment coefficient and the voltage stability adjustment coefficient obtained by pre-training in the database, adaptive correction of different elevator models, motor performance and on-site operating environment is realized. This method not only enhances the representation ability of the index to the real operating state, but also improves the consistency and robustness of cross-scene evaluation through the adjustment coefficient mechanism, providing precise and interpretable bottom driving dimension support for subsequent health judgment and fault analysis.

[0071] Specifically, the specific steps of analyzing the control response stability index of each operating stage of the elevator door machine are as follows: reading the average intensity of electromagnetic interference of the elevator door machine, the duration of electromagnetic interference of the elevator door machine, the average delay of control response, the maximum control response delay, and the control response delay fluctuation rate of each operating stage of the elevator door machine; comprehensively analyzing the average intensity of electromagnetic interference of the elevator door machine, the duration of electromagnetic interference of the elevator door machine, the average delay of control response, the maximum control response delay, and the control response delay fluctuation rate of each operating stage of the elevator door machine, respectively, to obtain the control response stability index of each operating stage of the elevator door machine.

[0072] The specific formula for calculating the control response stability index of a certain operating stage of the elevator door machine is as follows: ; wherein is the control response stability index of a certain operating stage of the elevator door machine, is the average delay of control response of a certain operating stage of the elevator door machine, is the control response delay fluctuation rate of a certain operating stage of the elevator door machine, is the average intensity of electromagnetic interference of a certain operating stage of the elevator door machine, is the duration of electromagnetic interference of a certain operating stage of the elevator door machine, is the maximum control response delay of a certain operating stage of the elevator door machine, is an offset angle adjustment factor stored in the database, used to prevent the denominator from being 0.

[0073] In this embodiment, by constructing the control response stability index, the control response average delay, maximum delay, delay fluctuation rate, and key control chain parameters such as the strength and duration of electromagnetic interference are fused and modeled, which can comprehensively evaluate the control chain stability and execution consistency of the elevator door machine in each running stage. At the same time, through the design of the denominator stability factor in the index formula, the calculation robustness under extreme values and special working conditions is improved. Compared with the traditional method which only relies on a single delay threshold, this index can reveal the dynamic response ability and internal coordination efficiency of the system under interference, especially for identifying implicit problems such as instruction-action asynchronization and control drift, which helps to improve the early warning ability and response accuracy of fault diagnosis.

[0074] Specifically, the specific steps of analyzing the door body structure posture index of each running stage of the elevator door machine are as follows: reading the door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency of each running stage of the elevator door machine; comprehensively analyzing the door body static friction load, door body dynamic friction load, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency of each running stage of the elevator door machine, respectively, to obtain the door body structure posture index of each running stage of the elevator door machine.

[0075] The specific formula for calculating the running health time sequence index of a running stage of the elevator door machine is as follows: ; wherein, is the door body structure posture index of a running stage of the elevator door machine, is the door body offset mutation rate of a running stage of the elevator door machine, is the door body jitter frequency of a running stage of the elevator door machine, is the door body offset average angle of a running stage of the elevator door machine, is an offset angle adjustment factor stored in the database, used to prevent the denominator from being 0, is the door body static friction load of a running stage of the elevator door machine, is the door body dynamic friction load of a running stage of the elevator door machine, is a friction load adjustment factor stored in the database, used to prevent the denominator from being 0, is the door body friction trend slope of a running stage of the elevator door machine.

[0076] The specific form of the tanh function is as follows: , wherein, is a natural constant and can be taken as 2.71 in the present embodiment, the domain is (-∞, +∞), and the range is (-1, +1).

[0077] In the present embodiment, by constructing a door body structure posture index, the door body static friction and dynamic friction load, friction trend slope, offset mutation rate, offset angle and jitter frequency and other key structure parameters are fused and modeled, the stability and posture consistency of the door body in the running process are comprehensively described, the tanh function is introduced in the formula for friction trend modeling, the abnormal value amplification effect can be effectively suppressed, the nonlinear response of gradual structure offset and wear risk is highlighted, and meanwhile, by setting the denominator stability factor, the numerical robustness of the index in the low-speed and low-friction scene is improved, the index shows higher sensitivity to hidden structure risks such as door rail deformation, component loosening and long-term wear, and helps to realize early perception and accurate judgment of structure faults.

[0078] Please refer to Figure 3 The embodiment of the present application provides a technical scheme: an elevator door machine system fault diagnosis system based on deep learning, comprising: a data acquisition unit configured to continuously acquire running state time series data of each running stage of the elevator door machine; a feature extraction unit configured to input the running state time series data of each running stage of the elevator door machine into a pre-trained deep learning model for feature extraction processing, to obtain running time series features of each running stage of the elevator door machine; a running health analysis unit configured to analyze running health time series indexes of each running stage of the elevator door machine based on the running time series features of each running stage of the elevator door machine; and a running fault analysis unit configured to judge and analyze the running health time series indexes of each running stage of the elevator door machine with respect to corresponding preset stage running health intervals, and to regard a running stage in which the running health time series index is outside the preset stage running health interval as a running fault.

[0079] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to these embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including all changes and modifications falling within the scope of the present application.

[0080] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the present application. Thus, if these modifications and changes of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and changes.

Claims

1. A method for fault diagnosis of elevator door system based on deep learning, characterized in that: The following steps are involved: Continuously obtain the operating status time series data of each operating stage of the elevator door machine; The operating status time series data of each operating stage of the elevator door machine are input into the pre-trained deep learning model for feature extraction processing to obtain the operating time series features of each operating stage of the elevator door machine; Based on the operation timing characteristics of each operation stage of the elevator door machine, the operation health timing index of each operation stage of the elevator door machine is analyzed; The operation health time series index of each operation stage of the elevator door machine is judged and analyzed with the corresponding preset stage operation health interval, and the operation stage where the operation health time series index is outside the preset stage operation health interval is regarded as an operation failure.

2. The elevator door system fault diagnosis method based on deep learning according to claim 1 is characterized in that: The operating status time series data includes the motor operating current, motor operating voltage, motor operating power, door machine electromagnetic interference intensity, control response delay, door body friction load, and door body offset angle at several time points.

3. The elevator door system fault diagnosis method based on deep learning according to claim 1 is characterized in that: The deep learning model includes an input embedding layer, a bidirectional GRU layer, a multi-head attention layer, a stage feature aggregation layer, and a feature output layer. The operating timing features include the peak value of the motor operating current, the variance of the motor operating current fluctuation, the number of high-frequency disturbances of the motor operating current, the stability coefficient of the motor operating voltage, the average intensity of the electromagnetic interference of the door machine, the duration of the electromagnetic interference of the door machine, the average delay of the control response, the maximum control response delay, the fluctuation rate of the control response delay, the static friction load of the door body, the dynamic friction load of the door body, the friction trend slope of the door body, the mutation rate of the door body offset, the average angle of the door body offset, and the door body jitter frequency.

4. The elevator door system fault diagnosis method based on deep learning according to claim 1 is characterized in that: The specific steps to obtain the operating sequence characteristics of each operating stage of the elevator door machine are as follows: In the input embedding layer of the deep learning model, the operating status time series data of each operating stage of the elevator door machine is received and normalized and feature dimension mapped to obtain the operating time series vector sequence of each operating stage of the elevator door machine with unified dimension; In the bidirectional GRU layer of the deep learning model, the sequence of operating time vectors for each operating stage of the elevator door machine is modeled, capturing the forward and backward dependencies of the operating states. The context state encoding at each time point is extracted to obtain the time series hidden state vector set for each operating stage of the elevator door machine. In the multi-head attention layer of the deep learning model, the time series hidden state vector set of each operating stage of the elevator door machine is evaluated for parameter importance and focused on abnormal time segments. The cross-feature representation of high-weight operating segments is extracted to obtain the stage operating state feature vector set of each operating stage of the elevator door machine with an attention weighting mechanism. In the stage feature aggregation layer of the deep learning model, the stage operation state feature vector set of each operation stage of the elevator door machine after attention weighting is globally averaged and aggregated to form a stage comprehensive representation vector corresponding to the operation stage; In the feature output layer of the deep learning model, the comprehensive representation vector of each operating stage of the elevator door machine is mapped and processed, and the operating sequence feature set of each operating stage of the elevator door machine is output.

5. The elevator door system fault diagnosis method based on deep learning according to claim 1 is characterized in that: The specific steps for analyzing the operation health time series index of each operation stage of the elevator door machine are as follows: Based on the operating sequence characteristics of each operating stage of the elevator door machine, the drive system health index, control response stability index, and door structure posture index of each operating stage of the elevator door machine are analyzed respectively; The drive system health index, control response stability index, and door structure posture index of each operating stage of the elevator door machine are comprehensively analyzed to obtain the operation health timing index of each operating stage of the elevator door machine.

6. The elevator door system fault diagnosis method based on deep learning according to claim 5 is characterized in that: The specific formula for calculating the operation health time series index of an elevator door machine at a certain operation stage is as follows: ； in, 、 、 、 The indexes are the operation health time series index, drive system health index, door structure posture index, and control response stability index of the elevator door machine at a certain operation stage. is a natural constant, 、 They are the structural health adjustment coefficient and the control coordination enhancement coefficient stored in the database respectively.

7. The elevator door system fault diagnosis method based on deep learning according to claim 5 is characterized in that: The specific steps for analyzing the drive system health index of the elevator door machine at each operating stage are as follows: Read the motor operating current peak value, motor operating current fluctuation variance, motor operating current high-frequency disturbance times, and motor operating voltage stability coefficient of each operating stage of the elevator door machine; A comprehensive analysis is performed on the motor operating current peak, motor operating current fluctuation variance, motor operating current high-frequency disturbance times, and motor operating voltage stability coefficient in each operating stage of the elevator door machine to obtain the drive system health index of the elevator door machine in each operating stage.

8. The elevator door system fault diagnosis method based on deep learning according to claim 5 is characterized in that: The specific steps for analyzing the control response stability index of each operating stage of the elevator door machine are as follows: Read the average electromagnetic interference intensity, duration of electromagnetic interference, average control response delay, maximum control response delay, and control response delay fluctuation rate of the elevator door machine at each operating stage; A comprehensive analysis is performed on the average electromagnetic interference intensity, duration of electromagnetic interference, average control response delay, maximum control response delay, and control response delay fluctuation rate of the elevator door machine in each operating stage to obtain the control response stability index of the elevator door machine in each operating stage.

9. The elevator door system fault diagnosis method based on deep learning according to claim 5, characterized in that: The specific steps for analyzing the door structure posture index of each operating stage of the elevator door machine are as follows: Read the static friction load, dynamic friction load, friction trend slope, displacement mutation rate, average displacement angle, and vibration frequency of the elevator door machine at each operating stage; The static friction load of the door body, dynamic friction load of the door body, door body friction trend slope, door body offset mutation rate, door body offset average angle, and door body jitter frequency of the elevator door machine in each operating stage are comprehensively analyzed to obtain the door body structure posture index of the elevator door machine in each operating stage.

10. A deep learning-based elevator door system fault diagnosis system, applying the deep learning-based elevator door system fault diagnosis method according to any one of claims 1 to 9, characterized in that: include: A data acquisition unit, used to continuously acquire the operating status time series data of each operating stage of the elevator door machine; A feature extraction unit is used to input the operating status time series data of each operating stage of the elevator door machine into the pre-trained deep learning model for feature extraction processing to obtain the operating time series features of each operating stage of the elevator door machine; An operation health analysis unit is used to analyze the operation health time series index of each operation stage of the elevator door machine based on the operation time series characteristics of each operation stage of the elevator door machine; The operation fault analysis unit is used to judge and analyze the operation health time series index of each operation stage of the elevator door machine and the corresponding preset stage operation health interval, and regard the operation stage where the operation health time series index is outside the preset stage operation health interval as an operation fault.

Images