Elevator hall door lock fault detection method, device and system

Abstract

The invention provides an elevator hall door lock fault detection method, device and system, and belongs to the technical field of elevator control. Comprising the following steps: S1, constructing an elevator hall door lock fault detection data set; s2, preprocessing the data and then extracting features; s3, constructing a fault detection model based on the features extracted in the step S2, and performing model training by using a cross entropy loss function to ensure that the model can accurately classify different fault states; s4, the model is trained based on dynamic learning rate adjustment and self-adaptive reweighting; and S5, deploying and using the model. Electrical signals and mechanical vibration signals of the elevator hall door lock are collected through the high-precision voltage and current sensor and the vibration sensor, fault detection and early warning operation are completed through the constructed fault detection model based on the deep neural network, adaptability to complex fault scenes is better, and detection accuracy and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator control, and in particular to a method, device and system for detecting elevator hall door lock faults. Background Art

[0002] Elevators, as an indispensable means of vertical transportation in modern buildings, have a safety and reliability that is directly related to the safety of people and property. Elevator hall door locks are crucial safety components in elevator systems. Their function is to ensure that the hall door remains closed during operation, preventing people or objects from accidentally falling into the hoistway. However, due to long-term use, mechanical wear, and electrical aging, elevator hall door locks may develop various faults, such as sticking, loosening, and poor electrical contact. If these faults are not discovered and addressed promptly, they can cause the elevator to stop operating, trapping people, or even lead to serious safety accidents.

[0003] The patent with application number 202122806504.0 provides an elevator door lock fault detection device and system, wherein the device includes at least one detection board, at least one outside call board and a main control board; each detection board is electrically connected to an outside call board and corresponds to the same elevator hall door, and all outside call boards are electrically connected to the main control board; each detection board is configured to detect the on and off status of the elevator door lock during the operation of the elevator car, and send a first fault signal to the corresponding outside call board when it detects that the elevator door lock is disconnected; each outside call board is configured to send a second fault signal to the main control board based on the received first fault signal; the main control board is configured to determine the fault position of the disconnected elevator door lock based on the received second fault signal, and send the fault position information to all outside call boards to control all outside call boards to display the fault position, so as to achieve rapid positioning of the elevator door lock fault position.

[0004] The patent with application number 202310317816.6 provides an elevator hall door lock fault detection method, device and system. After the elevator stops suddenly due to a hall door lock fault and the hall door lock fault information is displayed, in response to the opening operation of the hall door lock position detection function and the opening operation of the maintenance mode, a fault position acquisition signal is sent to the signal processing device to enable the signal processing device to detect whether the hall door lock of the corresponding floor is disconnected; after receiving the hall door lock disconnection signal of the target floor sent by the target signal processing device, a lighting outside call control signal is sent to the target signal processing device to light the outside call button of the target floor; after receiving the feedback signal returned by the target signal processing device that the lighting time of the outside call button reaches the specified time, the outside call stuck fault information is displayed, so that the staff can query according to the outside call stuck fault information to determine the fault floor corresponding to the hall door lock fault, which can facilitate the staff to quickly determine the fault floor corresponding to the hall door lock fault.

[0005] However, the above method has the following limitations:

[0006] 1. Existing technologies mainly rely on signal processing devices and simple logical judgments, such as the lighting duration of the hall call button. They lack the ability to intelligently analyze complex fault modes, and the level of intelligent detection methods is low.

[0007] 2. Existing methods only detect the disconnected state of the hall door lock and cannot cope with complex fault scenarios, such as partial failure of the hall door lock, intermittent faults, or multiple faults at once. Their adaptability to complex fault scenarios is limited.

[0008] 3. Existing methods do not fully utilize the potential value of fault data. For example, they do not analyze the frequency and time distribution of faults, cannot provide support for preventive maintenance of elevators, and lack in-depth analysis and utilization of fault data.

[0009] 4. Existing methods can only detect and locate faults after they occur, but cannot predict and prevent them. Although real-time data analysis can predict potential dangers and provide early warnings, it lacks support for fault prediction and prevention.

[0010] 5. The existing method still requires manual query and confirmation of the faulty floor after fault location, and fails to achieve a fully automated fault handling process. Although it can analyze data in real time and automatically issue warnings, it lacks automation and real-time performance. Summary of the Invention

[0011] The technical problem to be solved by the present invention is to provide an elevator hall door lock fault detection method, device and system to solve the problems of low intelligence level, limited adaptability to complex fault scenarios, and low detection accuracy and reliability of existing elevator hall door lock fault detection.

[0012] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0013] A method for detecting a fault in an elevator hall door lock comprises the following steps:

[0014] Step S1: Construct an elevator hall door lock fault detection dataset;

[0015] Step S2: extract features after preprocessing the data;

[0016] Step S3: Based on the features extracted in step S2, a fault detection model is constructed and trained using a cross-entropy loss function to ensure that the model can accurately classify different fault states.

[0017] Step S4: training the model based on dynamic learning rate adjustment and adaptive reweighting;

[0018] Step S5: Model deployment and use.

[0019] Preferably, the step S1: constructing an elevator hall door lock fault detection dataset comprises the following steps:

[0020] Step S1-1: Sensor deployment and data collection: Deploy high-precision voltage and current sensors and vibration sensors around the elevator hall door lock. The voltage and current sensors are used to collect electrical signal data, namely voltage and current signal data, in the door lock electrical circuit to capture electrical fault characteristics. The vibration sensor is used to collect vibration signal data during the door lock opening and closing process to capture abnormal vibration characteristics of mechanical components. The electrical and vibration signals of the door lock are collected to ensure that the data covers the normal state of the door lock and various fault states.

[0021] Step S1-2: Data labeling: Label the data collected in step S1-1. The labeling content includes the normal state and fault state of the door lock (such as electrical failure, mechanical jamming, looseness, etc.). Use the labeling tool to label the electrical signal and vibration signal to ensure that the label of each sample accurately reflects the actual state of the door lock. The labeled data set is recorded as where x i is the sensor data, y i is the corresponding label;

[0022] Step S1-3: Fault feature extraction and classification: Using prior knowledge and fault feature calculation formulas, fault features are extracted from the collected data;

[0023] For electrical signals, extract voltage fluctuations and current anomaly features:

[0024] F elec =[V std , I std , V max , I max ]

[0025] Among them, V std and I std are the standard deviations of voltage and current, V max and I max are the maximum values of voltage and current respectively;

[0026] For vibration signals, extract the vibration frequency and amplitude change characteristics;

[0027] F vib =[f mean , A max , f std ];

[0028] f mean is the mean value of the vibration frequency, A max is the maximum value of the amplitude, f stdis the standard deviation of the vibration frequency;

[0029] In the classification part, the data is classified according to the extracted features, and an elevator hall door lock fault detection dataset containing voltage, current, and vibration signals is constructed:

[0030]

[0031] and are the electrical characteristics and vibration characteristics of the i-th sample, y i For the corresponding label".

[0032] Preferably, the step S2: extracting features after preprocessing the data, comprises the following steps:

[0033] Step S2-1: Data preprocessing: First, a low-pass filter is used to remove high-frequency noise from the electrical signal and vibration signal;

[0034] V filtered =LowPassFilter(V raw , f c )

[0035] Among them, V raw is the original electrical signal or vibration signal, f c is the cutoff frequency;

[0036] Wavelet transform is then used to remove noise from the signal and retain valid information:

[0037] V denoised =WavcletDenoisc(V filtered ,threshold)

[0038] Among them, threshold is the denoising threshold;

[0039] Finally, the electrical signal and vibration signal are normalized to the range of [0,1] to eliminate the dimension difference:

[0040]

[0041] Among them, V min and V max are the minimum and maximum values of the signal respectively;

[0042] Step S2-2: Feature extraction: First, use Fourier transform to extract the frequency domain features of the electrical signal and capture the frequency distribution of the signal:

[0043] F freq =FFT(V normalized )

[0044] After frequency domain feature extraction, time domain feature extraction is performed. Wavelet transform is used to extract the time domain features of electrical and vibration signals and capture the temporal variation patterns of the signals:

[0045] F time =WavcletTransform(V normalized )

[0046] Different time series models are used for electrical signals and vibration signals respectively. LSTM is used for feature extraction of electrical signals, and GRU is used for feature extraction of vibration signals to generate high-dimensional electrical feature representation and high-dimensional vibration feature representation:

[0047] F etec =LSTM(V normalized )

[0048] F vib =GRU(V normalized ).

[0049] Preferably, the method of step three specifically includes:

[0050] Step S3-1: Multimodal feature fusion: First, it is necessary to extract and fuse the features of the electrical signal and the vibration signal:

[0051] The electrical features and vibration features are directly spliced into a comprehensive multimodal feature, which is expressed as follows:

[0052] F fusion =[F elec ; F vib ]

[0053] Among them, F fusion It is the fused multimodal feature;

[0054] Then, a multi-head attention mechanism is introduced to enhance the interaction modeling between features. The multi-head attention mechanism captures the potential correlation between electrical and vibration features by weighted interaction. The specific operation is as follows:

[0055] F fusion =MultiHcadAttention(F elec , F vib )

[0056] MultiHeadAttention is the operating function of the multi-head attention mechanism. In this way, the model can dynamically adjust the weights of different modal features to better integrate electrical and vibration information.

[0057] The output part is the fused multimodal feature F fusion, which contains comprehensive information of electrical and vibration signals, providing rich input for subsequent fault detection models;

[0058] Step S3-2: Fault detection model construction: After obtaining the fusion feature F fusion Finally, an efficient fault detection model needs to be built to classify different fault states. The core of the model is to further explore the complex relationship between multimodal features through a multi-head attention mechanism and graph neural network.

[0059] First, a multi-head attention mechanism is used to jointly model the fusion features:

[0060] F interact =MultiHeadAttention(F fusion )

[0061] Among them, F interact It is the feature representation after being processed by the multi-head attention mechanism;

[0062] This step can capture the local dependencies between multimodal features;

[0063] Next, we introduce graph neural networks to model the global dependencies between features. GNN represents features as nodes through a graph structure and updates node features through a message passing mechanism, as shown below:

[0064] F global =GNN(F interact )

[0065] Among them, F global It is the global feature representation after GNN processing;

[0066] Finally, the global features are mapped to the fault category space through the fully connected layer, which is expressed as follows:

[0067] F class =FC(F global )

[0068] Among them, FC represents the fully connected layer operation function, F class It is the feature representation mapped to the fault category space;

[0069] In order to obtain the probability distribution of the fault category, the Softmax function is used for calculation:

[0070] P=Softmax(F class )

[0071] Where P is the probability distribution of the fault class.

[0072] Preferably, the step S4 comprises the following steps based on the dynamic learning rate adjustment and adaptive reweighted training model:

[0073] Step S4-1: Adaptive reweighting strategy: During model training, in order to improve the model's ability to learn difficult samples, an adaptive reweighting strategy is introduced. This strategy dynamically adjusts the loss weight of samples, making the model pay more attention to difficult-to-classify samples.

[0074] The input part is the training data, including electrical characteristics, vibration characteristics and annotation labels. First, the cross entropy loss function is used to calculate the loss value of the training data, which is expressed as follows:

[0075] L=CrossEntropy(P,y)

[0076] Where P is the probability distribution of the fault category predicted by the model, and y is the true label;

[0077] For samples with poor training results (i.e., samples with larger loss values), a larger loss function value is assigned to them by calculating the adaptive weights:

[0078] w=α·L 2

[0079] Where w is the adaptive weight and α is the weight coefficient. In this way, difficult samples will be given higher weights during training, thereby improving the model's learning effect on these samples.

[0080] The output part is the weighted loss value L weighted , which is expressed as follows:

[0081] L weighted =w·L

[0082] Step S4-2: Dynamic learning rate adjustment: A dynamic learning rate adjustment strategy is introduced to optimize the training process. This strategy dynamically adjusts the learning rate according to the weighted loss value.

[0083] The input part is the weighted loss value L weighted , the learning rate adjustment formula is as follows:

[0084] μ * =μ·(1-L weighted ) 2 0.01

[0085] Where μ is the initial learning rate and μ* is the adjusted learning rate;

[0086] Step S4-3: Model training: After completing adaptive reweighting and dynamic learning rate adjustment, the model is trained using the Adam optimizer;

[0087] The input part is training data, including electrical characteristics, vibration characteristics and annotation labels. The update formula of the model parameters is as follows:

[0088]

[0089] Among them, θ is the model parameter, It is the gradient of the weighted loss value. By iteratively updating the model parameters, the trained model parameters θ* are finally obtained.

[0090] Preferably, the step S5: model deployment and use includes the following steps:

[0091] Step S5-1: Sensor deployment: High-precision voltage and current sensors and vibration sensors are deployed around the elevator hall door locks to fully capture the operating status of the door locks. These sensors can collect electrical and mechanical vibration signals from the door locks in real time, providing high-quality data support for subsequent fault detection.

[0092] Step S5-2: Data acquisition and transmission: Real-time acquisition of the door lock's electrical and vibration signals ensures the accuracy and robustness of the model, and ensures that the data covers the door lock's normal state and various fault states;

[0093] Step S5-3: Model deployment: After completing data collection and transmission, the elevator hall door lock fault detection model constructed in step S3 needs to be deployed to the computing terminal of the elevator system;

[0094] Step S5-4: Fault status determination: After the model is deployed, the system enters the real-time detection phase. The collected electrical and vibration signals are input into the fault detection model, which processes the input data in real time to predict the fault status of the door lock.

[0095] Step S5-5: Early warning and alarm: Set reasonable alarm thresholds to ensure the safety of the elevator system. Combined with the operating specifications and maintenance experience of the elevator system, early warning and alarm conditions can be set for different types of faults. For example, when an electrical fault or mechanical jam is detected, the system will trigger an alarm.

[0096] An elevator hall door lock fault detection device is applied to a computing terminal of an elevator system.

[0097] An elevator hall door lock fault detection system, the detection system comprising a sensor, a signal processor, a computing terminal, an elevator hall door lock, and an alarm module, the sensor being installed around the elevator hall door lock, the sensor, signal processor, computing terminal, and alarm module being communicatively connected, the detection system further comprising a storage device for storing fault detection results and related data, and the computing terminal for executing the elevator hall door lock fault detection method.

[0098] Compared with the prior art, the present invention has at least the following beneficial effects:

[0099] In this solution, high-precision voltage and current sensors and vibration sensors are used to collect electrical and mechanical vibration signals from elevator hall door locks in real time. Deep learning technology is then used to automatically extract fault characteristics, significantly reducing the need for manual intervention. Furthermore, the automated detection system can monitor the door lock status in real time, automatically identify the fault type, and promptly notify maintenance personnel through an alarm module, significantly improving the efficiency and accuracy of fault detection. Furthermore, the introduction of dynamic learning rate adjustment and an adaptive reweighting strategy enables the model to automatically adapt to different types of data distributions, thereby enhancing the automation level of elevator hall door lock fault detection.

[0100] In the above scheme, multimodal data fusion technology is used to integrate the characteristics of voltage, current, and vibration signals, making full use of information from multiple data sources. This can improve the intelligent level of fault detection. The multimodal feature fusion method can capture richer fault information. At the same time, the fault detection model based on the multi-head attention mechanism can effectively capture the complex relationships between multimodal features, further improving the intelligent analysis and judgment capabilities of fault modes and enhancing the intelligent level of hazard warning.

[0101] In the above scheme, by introducing deep learning models and multimodal data fusion technology, the accuracy of fault detection in complex scenarios can be significantly improved. The fault detection model can combine historical data and real-time monitoring data to effectively deal with complex fault scenarios. At the same time, multimodal data fusion and deep learning models can comprehensively consider complex factors, accurately identify fault types and locate fault locations, significantly improving the accuracy and reliability of detection and enhancing the accuracy of fault detection in complex scenarios.

[0102] In the above scheme, by using signal processing technology and deep learning models to extract multi-level features of voltage, current and vibration signals in the feature extraction stage, and by introducing dynamic learning rate adjustment and adaptive reweighting strategies in the model training stage, the model can automatically adapt to different types of data distributions, thereby improving the model's adaptability to complex scenarios and abnormal data. In addition, by expanding the training data set through data enhancement technology, the model's generalization ability is further improved, enabling it to maintain high detection accuracy in different elevator models and operating environments, thereby improving the generalization ability and robustness of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] The accompanying drawings, which are incorporated herein and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, further serve to explain the principles of the present disclosure and to enable one skilled in the relevant art to make and use the present disclosure.

[0104] Figure 1 This is a schematic diagram of the overall technical process structure of the elevator hall door lock fault detection method;

[0105] Figure 2 This is a schematic diagram of the structure of the voltage and current signal and vibration signal preprocessing and feature extraction module of the elevator hall door lock fault detection method;

[0106] Figure 3 A schematic diagram of a model structure diagram of an elevator hall door lock fault detection model of an elevator hall door lock fault detection method;

[0107] Figure 4 This is a schematic diagram of the actual flow chart structure of the model deployment of the elevator hall door lock fault detection method;

[0108] Figure 5 This is a schematic diagram of the overall system framework of the elevator hall door lock fault detection method.

[0109] As shown in the figure, in order to clearly implement the structure of the embodiment of the present invention, specific structures and devices are marked in the figure, but this is only for illustrative purposes and is not intended to limit the present invention to the specific structure, device and environment. According to specific needs, ordinary technicians in this field can adjust or modify these devices and environments, and the adjustments or modifications made are still included in the scope of the appended claims. DETAILED DESCRIPTION

[0110] The following describes in detail, with reference to the accompanying drawings and specific embodiments, a method, device, and system for detecting elevator hall door lock faults provided by the present invention. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative implementations for known techniques. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0111] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0112] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0113] It will be understood that the meanings of “on,” “over,” and “above” in this disclosure should be interpreted in the broadest manner, such that “on” means not only “directly on” something, but also includes being “on” something with intervening features or layers, and “on” or “over” means not only “on” or “above” something, but also includes being “on” or “above” something with no intervening features or layers.

[0114] Additionally, spatially relative terms such as "below," "beneath," "lower," "above," and "upper" may be used herein for descriptive convenience to describe the relationship of one element or feature to another element or features, as illustrated in the accompanying drawings. Spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the accompanying drawings. The device may be oriented in other ways, and the spatially relative descriptors used herein should be similarly interpreted accordingly.

[0115] like Figure 1-5 As shown, an embodiment of the present invention provides an elevator hall door lock fault detection method, comprising the following steps:

[0116] Step S1: Constructing an elevator hall door lock fault detection dataset: High-precision voltage and current sensors and vibration sensors are deployed around the elevator hall door locks to collect the voltage and current signals in the door lock electrical circuit and the vibration signals of the door lock during the opening and closing process. The collected data are annotated, including the normal state of the door lock and various fault states (such as electrical faults, mechanical jamming, looseness, etc.). Subsequently, prior knowledge and fault feature calculation formulas are used to extract and classify the fault features of the collected data, and an elevator hall door lock fault detection dataset containing voltage, current and vibration signals is constructed. This dataset provides high-quality multimodal data support for subsequent model training and verification.

[0117] The specific process includes:

[0118] Step S1-1: Sensor deployment and data collection: Elevator hall door lock fault detection dataset construction, which includes sensor deployment and data collection. High-precision voltage and current sensors and vibration sensors are deployed around the elevator hall door lock. The voltage and current sensors are used to collect electrical signal data in the door lock electrical circuit, namely voltage and current signal data, to capture electrical fault characteristics (such as short circuit, open circuit, poor contact, etc.). The vibration sensor is used to collect vibration signals during the door lock opening and closing process, capturing abnormal vibration characteristics of mechanical components (such as stuck, loose, etc.). The electrical and vibration signals of the door lock are collected to ensure that the data covers the normal state of the door lock and various fault states.

[0119] Step S1-2, data labeling: label the collected data. The labeling content includes the normal state and fault state of the door lock (such as electrical failure, mechanical stagnation, looseness, etc.). Use the labeling tool to label the electrical signal and vibration signal to ensure that the label of each sample accurately reflects the actual state of the door lock. The labeled data set is recorded as where x i is the sensor data, y i is the corresponding label;

[0120] Step S1-3, fault feature extraction and classification: Using prior knowledge and fault feature calculation formulas, perform fault feature extraction on the collected data. For electrical signals, extract voltage fluctuations and current anomaly features:

[0121] F eglec =[W std , I std , V max , I max ]

[0122] Among them, V std and I std are the standard deviations of voltage and current, V max and I max are the maximum values of voltage and current respectively;

[0123] For vibration signals, extract the vibration frequency and amplitude change characteristics;

[0124] F vib =[f mean , A max , f std ];

[0125] f mean is the mean value of the vibration frequency, A max is the maximum value of the amplitude, f std is the standard deviation of the vibration frequency;

[0126] In the classification part, the data is classified according to the extracted features, and an elevator hall door lock fault detection dataset containing voltage, current, and vibration signals is constructed:

[0127]

[0128] and are the electrical characteristics and vibration characteristics of the i-th sample, y i For the corresponding label.

[0129] Step S2, design of voltage, current signal and vibration signal preprocessing and feature extraction module: The present invention first preprocesses the signals collected by the voltage and current sensors, including filtering, denoising and normalization, to eliminate noise and dimensional differences, and then uses signal processing technology (such as Fourier transform, wavelet transform) to extract the frequency domain and time domain features of the voltage and current signals, and uses a time series model (such as LSTM, GRU) or a convolutional neural network (CNN) to extract features of the voltage and current signals to obtain a high-dimensional electrical feature representation. For the vibration signal, after similar preprocessing, the frequency domain and time domain features of the vibration signal are extracted using signal processing technology and a deep learning model to generate a high-dimensional vibration feature representation. Through the above steps, efficient feature extraction of the voltage, current and vibration signals is achieved;

[0130] The voltage, current, and vibration signal preprocessing and feature extraction module is designed. This process includes constructing an elevator hall door lock fault detection dataset as the input of S2, which contains electrical and vibration signals and their annotated labels.

[0131] The specific process includes the following steps:

[0132] Step S2-1: Data preprocessing: First, a low-pass filter is used to remove high-frequency noise from the electrical signal and vibration signal;

[0133] V filtered =LowPassFilter(V raw , f c )

[0134] Among them, V raw is the original electrical signal or vibration signal, f c is the cutoff frequency;

[0135] Wavelet transform is then used to remove noise from the signal and retain valid information:

[0136] V denoised =WavcletDenoisc(V filtered ,threshold)

[0137] Among them, threshold is the denoising threshold;

[0138] Finally, the electrical signal and vibration signal are normalized to the range of [0,1] to eliminate the dimension difference:

[0139]

[0140] Among them, V min and V max are the minimum and maximum values of the signal respectively;

[0141] Step S2-2: Feature extraction: First, use Fourier transform to extract the frequency domain features of the electrical signal and capture the frequency distribution of the signal:

[0142] F freq =FFT(V normalized )

[0143] After frequency domain feature extraction, time domain feature extraction is performed. Wavelet transform is used to extract the time domain features of electrical and vibration signals and capture the temporal variation patterns of the signals:

[0144] F time =WavcletTransform(V normalized )

[0145] Different time series models are used for electrical signals and vibration signals respectively. LSTM is used for feature extraction of electrical signals, and GRU is used for feature extraction of vibration signals to generate high-dimensional electrical feature representation and high-dimensional vibration feature representation:

[0146] F elec =LSTM(V normalized )

[0147] F vib =GRU(V normalized )

[0148] Step S2 preprocesses and extracts features from the electrical and vibration signals in the data set of step S1 to generate high-dimensional feature representations, including high-dimensional electrical feature representations F elec and high-dimensional vibrational feature representation F vib .

[0149] Step S3, multimodal feature fusion and fault detection model construction: Based on the electrical features and vibration features extracted in step S2, the present invention fuses the features of the two modes to construct a multimodal feature representation. Specifically, feature splicing or attention mechanism is used to realize the fusion of multimodal features to make full use of the complementary information between electrical signals and vibration signals. Subsequently, based on the fused multimodal features, a fault detection model based on a deep neural network is constructed. The model uses a multi-head attention mechanism or a graph neural network (GNN) to jointly model the multimodal features to capture the complex correlation between voltage and current signals and vibration signals. Finally, the fused features are input into the fully connected layer to predict the fault state of the door lock (such as normal, electrical fault, mechanical jamming, looseness, etc.), and the cross entropy loss function is used for model training to ensure that the model can accurately classify different fault states.

[0150] Multimodal feature fusion and fault detection model construction, the process includes: the model structure of the elevator hall door lock fault detection model is as follows Figure 3 As shown, the specific process includes the following steps:

[0151] Step S3-1: Multimodal feature fusion: Electrical signals and vibration signals are two important data sources for elevator hall door lock fault detection. The electrical signal reflects the electrical state of the door lock, such as changes in voltage and current, while the vibration signal captures the vibration characteristics of the door lock's mechanical motion. To fully utilize the information from these two signals, they must first be feature extracted and fused.

[0152] The input part includes the high-dimensional electrical features F extracted from the voltage and current signals elec and the high-dimensional vibration features F extracted from the vibration signal vib ,These features are usually extracted through signal processing techniques (such as Fourier ,transform, wavelet transform, or time-frequency analysis), which can ,effectively characterize the time and frequency domain characteristics of the ,signal;

[0153] Feature splicing is the first step in multimodal feature fusion.

[0154] The electrical features and vibration features are directly spliced into a comprehensive multimodal feature, which is expressed as follows:

[0155] F fusion =[F elec ; F vib ]

[0156] Among them, F fusion It is the fused multimodal feature. Although this simple splicing method is intuitive, it may not fully capture the complex relationship between the two modalities. Therefore, the multi-head attention mechanism is further introduced to enhance the interaction modeling between features.

[0157] The multi-head attention mechanism captures the potential correlation between electrical and vibration features by performing weighted interactions between them. The specific operation is as follows:

[0158] F fusion =MultiHcadAttention(F elec , F vib )

[0159] Among them, MultiHeadAttention is the operating function of the multi-head attention mechanism; in this way, the model can dynamically adjust the weights of different modal features to better integrate electrical and vibration information.

[0160] The output part is the fused multimodal feature F fusion , which contains comprehensive information of electrical and vibration signals, providing rich input for subsequent fault detection models;

[0161] D3. Fault detection model construction: After obtaining the fusion feature F fusion Finally, an efficient fault detection model needs to be built to classify different fault states. The core of the model is to further explore the complex relationship between multimodal features through a multi-head attention mechanism and graph neural network (GNN).

[0162] First, a multi-head attention mechanism is used to jointly model the fusion features:

[0163] F interact =MultiHeadAttention(F fusion )

[0164] Among them, F interact It is the feature representation after being processed by the multi-head attention mechanism;

[0165] This step can capture the local dependencies between multimodal features;

[0166] Next, we introduce a graph neural network (GNN) to model the global dependencies between features. GNN represents features as nodes through a graph structure and updates node features through a message passing mechanism, as shown below:

[0167] F global =GNN(F interact )

[0168] Among them, F global It is the global feature representation after GNN processing; the advantage of GNN is that it can capture the nonlinear relationship between features, thereby improving the expressive power of the model;

[0169] Finally, the global features are mapped to the fault category space through the fully connected layer:

[0170] F class =FC(F global )

[0171] Among them, FC represents the fully connected layer operation function, F class It is the feature representation mapped to the fault category space;

[0172] In order to obtain the probability distribution of the fault category, the Softmax function is used for calculation:

[0173] P=Softmax(F class )

[0174] Among them, P is the probability distribution of fault categories, and the output of the model includes multiple fault categories, such as normal, electrical failure, mechanical jamming, looseness, etc.

[0175] Step S4: Constructing a training strategy based on dynamic learning rate adjustment and adaptive reweighting: During model training, for training data with poor training results, a larger loss function value is assigned to it by calculating adaptive weights, so as to help the model pay more attention to sample data with poor training results, thereby allowing the model to maintain its focus on challenging samples, thereby enhancing the overall performance of the model. By dynamically adjusting the learning rate during model training, the generalization ability of the model is ensured;

[0176] A training strategy based on dynamic learning rate adjustment and adaptive reweighting is constructed. Specifically, the model output (fault category probability distribution) in step S3 is used as the input of step S4 to calculate the loss value and adjust the learning rate. Through the adaptive reweighting strategy and dynamic learning rate adjustment strategy of step S4, the model can learn multimodal features more efficiently, thereby improving the accuracy of fault detection.

[0177] The specific process includes the following steps:

[0178] Step S4-1: Adaptive reweighting strategy: During the model training process, the difficulty and importance of different samples may be different. In order to improve the model's learning ability for difficult samples, an adaptive reweighting strategy is introduced. This strategy dynamically adjusts the loss weight of the sample to make the model pay more attention to difficult-to-classify samples.

[0179] The input part is the training data, including electrical characteristics, vibration characteristics and annotation labels. First, the cross entropy loss function is used to calculate the loss value of the training data, which is expressed as follows:

[0180] L=CrossEntropy(P,y)

[0181] Where P is the probability distribution of the fault category predicted by the model, and y is the true label;

[0182] For samples with poor training results (i.e., samples with larger loss values), a larger loss function value is assigned to them by calculating the adaptive weights:

[0183] w=α·L 2

[0184] Where w is the adaptive weight and α is the weight coefficient. In this way, difficult samples will be given higher weights during training, thereby improving the model's learning effect on these samples.

[0185] The output part is the weighted loss value L weighted , which is expressed as follows:

[0186] L weighted =w·L;

[0187] Step S4-2: Dynamic learning rate adjustment: The learning rate is a key hyperparameter in model training. To optimize the training process, a dynamic learning rate adjustment strategy is introduced. This strategy dynamically adjusts the learning rate according to the weighted loss value, so as to achieve rapid convergence in the early stage of training and fine-tune it in the later stage of training.

[0188] The input part is the weighted loss value L weighted , the learning rate adjustment formula is as follows:

[0189] μ * =μ·(1-L weighted ) 2 0.01

[0190] Here, μ is the initial learning rate and μ* is the adjusted learning rate. In this way, the learning rate will be dynamically adjusted as the loss value changes, thereby improving the training efficiency of the model.

[0191] Step S4-3: Model training: After completing adaptive reweighting and dynamic learning rate adjustment, the model is trained using the Adam optimizer. The Adam optimizer combines the advantages of momentum and adaptive learning rate, and can converge quickly during training.

[0192] The input part is training data, including electrical characteristics, vibration characteristics and annotation labels. The update formula of the model parameters is as follows:

[0193]

[0194] Among them, θ is the model parameter, It is the gradient of the weighted loss value. By iteratively updating the model parameters, the trained model parameters θ* are finally obtained.

[0195] To ensure that the fault detection model constructed in step S3 is closely integrated with the training strategy in step S4, the present invention provides an efficient and accurate solution for elevator hall door lock fault detection through multimodal feature fusion and fault detection model construction in step S3, as well as adaptive reweighting and dynamic learning rate adjustment in step S4.

[0196] Step S5, actual deployment and use of the model: Deploy voltage and current sensors and vibration sensors around the elevator hall door lock to collect voltage and current signals and vibration signals in real time. The collected data is fed into the elevator hall door lock fault detection model. The model processes the input data in real time, predicts the fault status of the door lock, and combines it with the alarm threshold of the elevator system to achieve real-time early warning and alarm operation of elevator hall door lock faults. Through the deployment and use of this system, the safety of elevator operation can be effectively ensured and safety accidents caused by door lock failures can be reduced.

[0197] The fault detection model in step S5 is based on the multimodal feature fusion and fault detection model constructed in step S3. By fusing electrical features and vibration features, the model can fully capture the operating status of the door lock, thereby improving the accuracy of fault detection. The model deployment and real-time detection in step S5 rely on the adaptive reweighting and dynamic learning rate adjustment strategies in step S4. These strategies ensure the stability and robustness of the model during training, thereby performing well in actual deployment. Finally, the real-time detection results of step S5 will be fed back to the alarm module of the elevator system, triggering early warning and alarm operations. This closed-loop design ensures the safety of elevator operation while also providing maintenance personnel with efficient tools and support. Its specific implementation steps include:

[0198] Step S5-1: Sensor deployment: In the elevator hall door lock fault detection system, sensor deployment is a key step in achieving real-time data collection. To fully capture the operating status of the door lock, high-precision voltage and current sensors and vibration sensors need to be deployed around the elevator hall door lock. These sensors can collect the door lock's electrical signals (voltage and current signals) and mechanical vibration signals in real time, providing high-quality data support for subsequent fault detection.

[0199] Voltage and current sensors are used to collect voltage and current signals in the door lock's electrical circuit. These signals can reflect the door lock's electrical status, such as fault characteristics such as short circuit, open circuit, and poor contact. By monitoring voltage and current changes in real time, early signs of electrical faults can be detected in time to prevent further deterioration of the fault.

[0200] The vibration sensor is used to collect vibration signals during the door lock opening and closing process. The mechanical components of the door lock (such as the lock tongue, spring, etc.) will produce specific vibration patterns during operation. When the mechanical components become stuck, loose or other abnormalities occur, the frequency and amplitude of the vibration signal will change. By analyzing these changes, mechanical faults can be accurately identified.

[0201] Step S5-2: Data Collection and Transmission: Data collection is one of the core components of the fault detection system. To ensure the accuracy and robustness of the model, it is necessary to collect the electrical and vibration signals of the door lock in real time, and ensure that the data covers the normal state of the door lock and various fault states. Through high-frequency data collection, subtle changes in the door lock operation can be captured, thereby improving the sensitivity of fault detection;

[0202] Step S5-3: Model deployment: After completing data collection and transmission, the elevator hall door lock fault detection model constructed in step S3 needs to be deployed to the computing terminal of the elevator system. To achieve efficient real-time detection, the model is usually deployed on an edge computing device;

[0203] Step S5-4: Fault status determination: After the model is deployed, the system enters the real-time detection phase. The collected electrical and vibration signals are input into the fault detection model. The model processes the input data in real time and predicts the fault status of the door lock. Fault status includes normal, electrical fault, mechanical jamming, looseness, and other types.

[0204] Step S5-5: Early Warning and Alarm: To ensure the safety of the elevator system, reasonable alarm thresholds need to be set. Based on the elevator system's operating specifications and maintenance experience, early warning and alarm conditions can be set for different types of faults. For example, when an electrical fault or mechanical jam is detected, the system will trigger an alarm.

[0205] The present invention also provides an elevator hall door lock fault detection device, which is applied to a computing terminal of an elevator system.

[0206] The present invention also provides an elevator hall door lock fault detection system, the detection system comprising a sensor, a signal processor, a computing terminal, an elevator hall door lock, and an alarm module, the sensor being installed around the elevator hall door lock, the sensor, signal processor, computing terminal, and alarm module being communicatively connected, and the computing terminal being used to execute the elevator hall door lock fault detection method;

[0207] The detection system also includes a storage device for storing fault detection results and related data to facilitate subsequent data analysis and system optimization. The stored data can be used to analyze the frequency and trend of faults and provide a basis for preventive maintenance of elevators.

[0208] The entire system is integrated and coordinated through the elevator hall door fault detection module to ensure seamless connection of various components and processes, realize real-time monitoring and early warning of elevator hall door lock failures, thereby improving the safety and reliability of elevator operation.

[0209] In the present invention, the use process of the elevator hall door lock fault detection system is as follows:

[0210] First, high-precision voltage and current sensors and vibration sensors are deployed around the elevator hall door locks. These sensors collect the door lock's electrical signals (voltage and current) and mechanical vibration signals in real time. The collected signals are transmitted to memory via the signal input module for subsequent processing. Simultaneously, the power supply module provides stable power support for the entire system, ensuring the normal operation of all components.

[0211] The signal processing unit preprocesses the voltage, current, and vibration signals in memory, including filtering, denoising, and normalization, to improve signal quality and extract useful features. These features are then fed into the fault detection model, which, based on a deep learning algorithm, can identify and analyze fault characteristics in the signals.

[0212] After the fault detection model processes the signal, the fault status perception module will determine whether there is a door lock fault based on the output results of the model. If a fault is detected, the elevator hall door fault detection system terminal will generate a corresponding alarm signal and issue an early warning through the elevator hall door fault early warning module to remind maintenance personnel to inspect and repair.

[0213] Conduct experimental verification analysis on the invention

[0214] This experiment is based on the actual deployment of an elevator hall door lock fault detection system. Over a 30-day period, operating data from the elevator hall door locks was continuously collected. This data included four multi-source time series data items: voltage (V), current (I), vibration amplitude (A), and vibration frequency (f), as well as four target parameters: fault status (e.g., normal, electrical fault, mechanical jamming, and looseness). The sampling frequency of the data acquisition system was set to 20 Hz, with a training to test data ratio of 8:2.

[0215] In terms of model performance evaluation, the model was verified from three perspectives: fault classification accuracy, anomaly detection accuracy, and real-time detection latency. For the fault classification task, the following three error evaluation indicators were used for analysis:

[0216] 1. Mean Absolute Error (MAE): This measures the average absolute error between the predicted fault category and the actual fault category. A smaller value indicates a more accurate classification result.

[0217] 2. Mean Squared Error (MSE): Calculates the average of the squared errors between the predicted value and the true value. This can amplify the impact of larger errors on the overall error and assess the stability of the model.

[0218] 3. Root Mean Squared Error (RMSE): This is the square root of the MSE, which measures the overall level of prediction error. The unit is consistent with the original data and more intuitively reflects the classification accuracy.

[0219] The experimental results are shown in Table 1.

[0220] Table 1. Fault classification experimental results

[0221]

[0222]

[0223] For the anomaly detection task, accuracy is used as the core evaluation indicator, and the experimental results are shown in Table 2. The model can effectively identify abnormal data in electrical and vibration signals, avoiding the interference of abnormal data on fault detection results.

[0224] Table 2 Experimental results of anomaly detection accuracy

[0225] Exception data types Abnormality judgment accuracy (%) Voltage V 93.6 Current I 92.9 Vibration amplitude A 95.1 Vibration frequency f 95.6

[0226] For the real-time delay detection task, the average response time (ART) is used as the evaluation indicator, and the experimental results are shown in Table 3. A lower ART indicates that the model can complete fault detection in a short time, meeting the real-time requirements of the elevator system.

[0227] Table 3. Real-time detection delay experimental results

[0228] Detection tasks Average response time (ms) Electrical fault detection 13.65 Mechanical fault detection 16.54 Comprehensive fault detection 14.87

[0229] Experimental results show that the method of the present invention achieves low MAE, MSE and RMSE values in the fault classification task, proving that the model can accurately identify the fault status of the elevator hall door lock; in the anomaly detection task, the model has a high accuracy rate and effectively avoids abnormal data interference; in the real-time detection task, the average response time is short, meeting the real-time requirements of the elevator system. In summary, the experiment verifies the effectiveness and practical application value of the present invention in the elevator hall door lock fault detection task.

[0230] Compared with the prior art, the present invention has the following beneficial effects:

[0231] (1) Improve the automation level of elevator hall door lock fault detection: Use high-precision voltage and current sensors and vibration sensors to collect the electrical signals and mechanical vibration signals of the elevator hall door lock in real time, and use deep learning technology to automatically extract fault features, which can significantly reduce the need for manual intervention. Relying on manual inspections and simple signal processing is inefficient and prone to missed detection. The automated detection system can monitor the door lock status in real time, automatically identify the fault type (such as electrical fault, mechanical jamming, looseness, etc.), and promptly notify maintenance personnel through the alarm module, greatly improving the efficiency and accuracy of fault detection. In addition, the introduction of dynamic learning rate adjustment and adaptive reweighting strategy enables the model to automatically adapt to different types of data distribution;

[0232] (2) Enhance the intelligence level of danger warning: Through multimodal data fusion technology, the characteristics of voltage, current and vibration signals are integrated, and the information of multiple data sources is fully utilized to improve the intelligence level of fault detection. Relying only on a single type of data (such as electrical signals or simple mechanical signals) cannot fully reflect the fault status of the door lock. The multimodal feature fusion method can capture richer fault information, such as identifying poor electrical contact through voltage and current signals, and detecting mechanical jamming through vibration signals. The fault detection model based on the multi-head attention mechanism can effectively capture the complex relationship between multimodal features, further improving the intelligent analysis and judgment capabilities of fault modes;

[0233] (3) Enhance the accuracy of fault detection in complex scenarios: By introducing deep learning models and multimodal data fusion technology, the accuracy of fault detection in complex scenarios can be significantly improved. The fault detection model proposed by the present invention can combine historical data and real-time monitoring data to effectively deal with complex fault scenarios. For example, during the operation of an elevator, the door lock may have various faults due to mechanical wear, electrical aging or external environmental factors (such as temperature and humidity changes). The multimodal data fusion and deep learning model of the present invention can comprehensively consider these factors, accurately identify the fault type and locate the fault location, significantly improving the accuracy and reliability of detection;

[0234] (4) Improving the generalization ability and robustness of fault detection: The present invention adopts a variety of technical means to improve the generalization ability and robustness of the system during the feature extraction and model training process. For example, in the feature extraction stage, multi-level features of voltage, current and vibration signals are extracted through signal processing technology and deep learning models. In the model training stage, the introduction of dynamic learning rate adjustment and adaptive reweighting strategy enables the model to automatically adapt to different types of data distribution, thereby improving the model's adaptability to complex scenarios and abnormal data. In addition, the training data set is expanded through data enhancement technology, which further improves the generalization ability of the model, so that it can maintain a high detection accuracy under different elevator models and operating environments.

[0235] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0236] Those skilled in the art will understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc.

[0237] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for detecting a fault in an elevator hall door lock, characterized in that: The following steps are involved: Step S1: Construct an elevator hall door lock fault detection dataset; Step S2: extract features after preprocessing the data; Step S3: Based on the features extracted in step S2, a fault detection model is constructed and trained using a cross-entropy loss function to ensure that the model can accurately classify different fault states. Step S4: training the model based on dynamic learning rate adjustment and adaptive reweighting; Step S5: Model deployment and use.

2. The elevator hall door lock fault detection method according to claim 1, characterized in that: The step S1: constructing an elevator hall door lock fault detection dataset includes the following steps: Step S1-1: Sensor deployment and data collection: Deploy high-precision voltage and current sensors and vibration sensors around the elevator hall door lock. The voltage and current sensors are used to collect electrical signal data, namely voltage and current signal data, in the door lock electrical circuit to capture electrical fault characteristics. The vibration sensor is used to collect vibration signal data during the door lock opening and closing process to capture abnormal vibration characteristics of mechanical components. The electrical and vibration signals of the door lock are collected to ensure that the data covers the normal state of the door lock and various fault states. Step S1-2: Data labeling: Label the data collected in step S1-1. The labeling content includes the normal state and fault state of the door lock (such as electrical failure, mechanical jamming, looseness, etc.). Use the labeling tool to label the electrical signal and vibration signal to ensure that the label of each sample accurately reflects the actual state of the door lock. The labeled data set is recorded as where x i is the sensor data, y i is the corresponding label; Step S1-3: Fault feature extraction and classification: Using prior knowledge and fault feature calculation formulas, fault features are extracted from the collected data; For electrical signals, extract voltage fluctuations and current anomaly features: F elec =[V std ,I std ,V max ,I max ] Among them, V std and I std are the standard deviations of voltage and current, V max and I max are the maximum values of voltage and current respectively; For vibration signals, extract the vibration frequency and amplitude change characteristics; F vib =[f mean ,f max ,f std ]; f mean is the mean value of the vibration frequency, A max is the maximum value of the amplitude, f std is the standard deviation of the vibration frequency; In the classification part, the data is classified according to the extracted features, and an elevator hall door lock fault detection dataset containing voltage, current, and vibration signals is constructed: and are the electrical characteristics and vibration characteristics of the i-th sample, y i For the corresponding label.

3. The elevator hall door lock fault detection method according to claim 2, characterized in that: The step S2: extracting features after preprocessing the data, includes the following steps: Step S2-1: Data preprocessing: First, a low-pass filter is used to remove high-frequency noise from the electrical signal and vibration signal; V filtered =LowPassFilter(V raw ,f c ) Among them, V raw is the original electrical signal or vibration signal, f c is the cutoff frequency; Wavelet transform is then used to remove noise from the signal and retain valid information: V denoised =WavcleTDenoisc(V filtered ,threshold) Among them, threshold is the denoising threshold; Finally, the electrical signal and vibration signal are normalized to the range of [0,1] to eliminate the dimension difference: Among them, V min and V max are the minimum and maximum values of the signal respectively; Step S2-2: Feature extraction: First, use Fourier transform to extract the frequency domain features of the electrical signal and capture the frequency distribution of the signal: F freq =FFT(V normalized ) After frequency domain feature extraction, time domain feature extraction is performed, and wavelet transform is used to extract the time domain features of electrical and vibration signals to capture the time variation patterns of the signals: F time =WavcletTransform(V normalized ) Different time series models are used for electrical signals and vibration signals respectively. LSTM is used for feature extraction of electrical signals, and GRU is used for feature extraction of vibration signals to generate high-dimensional electrical feature representation and high-dimensional vibration feature representation: F elec =LSTM(V normalized ) F vib =GRU(V normalized )。 4. The elevator hall door lock fault detection method according to claim 3, characterized in that: The method of step three specifically includes: Step S3-1: Multimodal feature fusion: First, the electrical signal and vibration signal need to be feature extracted and fused; The electrical features and vibration features are directly spliced into a comprehensive multimodal feature, which is expressed as follows: F fusion =[F elec ;F vib ] Among them, F fusion It is the fused multimodal feature; Then, a multi-head attention mechanism is introduced to enhance the interaction modeling between features. The multi-head attention mechanism captures the potential correlation between electrical and vibration features by weighted interaction. The specific operation is as follows: F fusion =MultiHcadAttention(F elec ,F vib ) MultiHeadAttention is the operating function of the multi-head attention mechanism. In this way, the model can dynamically adjust the weights of different modal features to better integrate electrical and vibration information. The output part is the fused multimodal feature F fusion , which contains comprehensive information of electrical and vibration signals, providing rich input for subsequent fault detection models; Step S3-2: Fault detection model construction: After obtaining the fusion feature F fusion Finally, an efficient fault detection model needs to be built to classify different fault states. The core of the model is to further explore the complex relationship between multimodal features through a multi-head attention mechanism and graph neural network. First, a multi-head attention mechanism is used to jointly model the fusion features: F interact =MultiHeadAttentionn(F fusion ) Among them, F interact It is the feature representation after being processed by the multi-head attention mechanism; This step can capture the local dependencies between multimodal features; Next, we introduce graph neural networks to model the global dependencies between features. GNN represents features as nodes through a graph structure and updates node features through a message passing mechanism, as shown below: F global =GNN(F interact ) Among them, F global It is the global feature representation after GNN processing; Finally, the global features are mapped to the fault category space through the fully connected layer, which is expressed as follows: F class =FC(F global ) Among them, FC represents the fully connected layer operation function, F class It is the feature representation mapped to the fault category space; In order to obtain the probability distribution of the fault category, the Softmax function is used for calculation: P=Softmax(F class ) Where P is the probability distribution of the fault class.

5. The elevator hall door lock fault detection method according to claim 4, characterized in that: The step S4 includes the following steps based on dynamic learning rate adjustment and adaptive reweighting training model: Step S4-1: Adaptive reweighting strategy: During model training, in order to improve the model's ability to learn difficult samples, an adaptive reweighting strategy is introduced. This strategy dynamically adjusts the loss weight of samples to make the model pay more attention to difficult-to-classify samples. The input part is the training data, including electrical characteristics, vibration characteristics and annotation labels. First, the cross entropy loss function is used to calculate the loss value of the training data, which is expressed as follows: L=CrossEntropy(P,y) Where P is the probability distribution of the fault category predicted by the model, and y is the true label; For samples with poor training results (i.e., samples with larger loss values), a larger loss function value is assigned to them by calculating the adaptive weights: w=α·L 2 Where w is the adaptive weight and α is the weight coefficient. In this way, difficult samples will be given higher weights during training, thereby improving the model's learning effect on these samples. The output part is the weighted loss value L weighted , which is expressed as follows: L weighted =w·L Step S4-2: Dynamic learning rate adjustment: A dynamic learning rate adjustment strategy is introduced to optimize the training process. This strategy dynamically adjusts the learning rate according to the weighted loss value. The input part is the weighted loss value L weighted , the learning rate adjustment formula is as follows: m * =μ·(1-L weighted ) 2 ·0.01 Among them, μ is the initial learning rate, μ * is the adjusted learning rate; Step S4-3: Model training: After completing adaptive reweighting and dynamic learning rate adjustment, the model is trained using the Adam optimizer; The input part is training data, including electrical characteristics, vibration characteristics and annotation labels. The update formula of the model parameters is as follows: Among them, θ is the model parameter, It is the gradient of the weighted loss value. By iteratively updating the model parameters, the trained model parameters θ* are finally obtained.

6. The elevator hall door lock fault detection method according to claim 5, characterized in that: The step S5: model deployment and use includes the following steps: Step S5-1: Sensor deployment: High-precision voltage and current sensors and vibration sensors are deployed around the elevator hall door locks to fully capture the operating status of the door locks. These sensors can collect electrical and mechanical vibration signals from the door locks in real time, providing high-quality data support for subsequent fault detection. Step S5-2: Data acquisition and transmission: Real-time acquisition of the door lock's electrical and vibration signals ensures the accuracy and robustness of the model, and ensures that the data covers the door lock's normal state and various fault states; Step S5-3: Model deployment: After completing data collection and transmission, the elevator hall door lock fault detection model constructed in step S3 needs to be deployed to the computing terminal of the elevator system; Step S5-4: Fault status determination: After the model is deployed, the system enters the real-time detection phase. The collected electrical and vibration signals are input into the fault detection model, which processes the input data in real time to predict the fault status of the door lock. Step S5-5: Early warning and alarm: Set reasonable alarm thresholds to ensure the safety of the elevator system. Combined with the operating specifications and maintenance experience of the elevator system, early warning and alarm conditions can be set for different types of faults. For example, when an electrical fault or mechanical jam is detected, the system will trigger an alarm.

7. An elevator hall door lock fault detection device, characterized in that: The detection device is applied to a computing terminal of an elevator system.

8. An elevator hall door lock fault detection system, characterized in that: The detection system includes a sensor, a signal processor, a computing terminal, an elevator hall door lock and an alarm module. The sensor is installed around the elevator hall door lock. The sensor, signal processor, computing terminal and alarm module are communicatively connected. The detection system also includes a storage device for storing fault detection results and related data. The computing terminal is used to execute the elevator hall door lock fault detection method according to any one of claims 1 to 6.

Images