Elevator abnormal state online monitoring method and system based on vibration signal analysis

Abstract

The invention provides an elevator abnormal state on-line monitoring method and system based on vibration signal analysis, and the method comprises the steps: introducing a deep residual shrinkage network DRSN-CW, and replacing a convolutional layer in the DRSN-CW network with a deep separable convolutional layer, thereby obtaining a DS-DRSN-CW network, and carrying out the online monitoring of the abnormal state of an elevator through the DS-DRSN-CW network. According to the network, the model calculation complexity is effectively reduced, and meanwhile, high-precision extraction and recognition of fault features are considered. On the system level, a distributed framework based on edge computing is constructed, and real-time collection and wireless transmission of original vibration signals of the elevator are achieved through sensor nodes and a communication module; at the same time, the MCU unit embedded in the DS-DRSN-CW network is used to complete fault identification, and only key data is uploaded to the cloud. Through synchronous optimization of the method and the system, real-time sensing of elevator fault signals is achieved, the accuracy is guaranteed, meanwhile, a lightweight reliable system is constructed, and an efficient solution is provided for intelligent operation and maintenance of the elevator.

Description

Technical Field

[0001] This invention relates to the field of elevator fault monitoring technology, specifically to an online monitoring method and system for abnormal elevator conditions based on vibration signal analysis. Background Technology

[0002] As a core piece of equipment in urban vertical transportation, ensuring the safe and stable operation of elevators is crucial. However, during long-term, high-frequency use, heavy-duty elevators can generate abnormal micro-vibration signals due to factors such as dynamic load changes, fatigue wear of mechanical components, and track misalignment. These signals are often early signs of major malfunctions. Therefore, achieving rapid fault location and accurate early warning for elevators has become a key technical challenge for ensuring public safety and reducing operation and maintenance costs.

[0003] In elevator vibration signal processing, traditional methods such as wavelet decomposition and wavelet packet decomposition are not effective for analyzing vibration signals under complex operating conditions. Environmental noise, electromagnetic interference, and other factors accompanying elevator operation can easily cause signal distortion, leading to errors in fault feature extraction and consequently affecting the accurate determination of elevator operating conditions.

[0004] In terms of elevator fault detection technology, existing technologies are mainly divided into two categories: offline analysis and online monitoring. Offline data analysis relies on post-event sampling and cannot capture transient fault characteristics in real time, making it difficult to meet proactive maintenance needs. In contrast, while online monitoring technology can achieve real-time perception of operating status, it still faces significant technical bottlenecks. On the one hand, traditional wired monitoring solutions require laying a large number of cables in complex environments such as the car, hoistway, and machine room, resulting in high construction difficulty, poor scalability, and high maintenance costs. On the other hand, although emerging wireless online monitoring technology has become the mainstream due to its advantages of flexible deployment and convenient installation, wireless sensor nodes generally suffer from weak computing power and poor battery life, making it difficult to complete long-term and complex signal denoising and feature extraction tasks at the edge. Summary of the Invention

[0005] Purpose of the invention: The first purpose of this invention is to provide a low-computational-requirement, high-precision online monitoring method for elevator abnormal conditions based on vibration signal analysis; the second purpose is to provide an online monitoring system for elevator abnormal conditions based on vibration signal analysis.

[0006] Technical solution: An online monitoring method for abnormal elevator conditions based on vibration signal analysis, comprising the following steps:

[0007] S1. Collect the original vibration signal of the elevator;

[0008] S2, on the basis of the DRSN-CW network, the deep separable convolution layer is used to replace the convolution layer in the DRSN-CW network, and a DS-DRSN-CW network is obtained, and the trained DS-DRSN-CW network is used for feature extraction and fault identification of the original elevator vibration signal;

[0009] S3, when the DS-DRSN-CW network identifies an abnormal vibration signal, the abnormal vibration signal and the fault identification result are uploaded to the cloud data center through wireless communication.

[0010] Specifically, the DS-DRSN-CW network comprises an input layer, a deep separable convolution layer, a plurality of DS-RSBU-CW layers, a global average pooling layer and a full connection layer connected in sequence, and the DS-RSBU-CW layer is constructed by replacing the convolution layer of the RSBU-CW module in the DRSN-CW network with a deep separable convolution layer.

[0011] Specifically, the DS-RSBU-CW layer obtains channel statistical information through global average pooling, and then generates a channel threshold value through two layers of full connection network and a Sigmoid activation function, and designs a soft threshold function using the channel threshold value for screening.

[0012] Specifically, the soft threshold function formula is:

[0013]

[0014] In the formula, y is the output feature, x is the input feature, and τ is the channel threshold value.

[0015] Specifically, the channel threshold value is obtained by the following steps:

[0016] The feature matrix of the channel is globally averaged by using global average pooling to obtain a scalar representing the overall feature strength of the channel, and the scalar formula is:

[0017]

[0018] In the formula, S C is the scalar of the channel C, X is the channel feature matrix, H is the height, and W is the width;

[0019] The scalar is nonlinearly transformed and dimensionally adjusted by using a full connection network, and then the result is mapped to the interval (0, 1) by using a Sigmoid activation function, and the channel threshold value corresponding to the channel is output in combination with the amplitude range of the channel feature.

[0020] Specifically, step S3 comprises:

[0021] The fault identification result output by the DS-DRSN-CW network is divided into two states of normal and abnormal, if the output is abnormal, the abnormal vibration signal and the fault identification result are uploaded to the cloud data center through wireless communication, if the output is normal, the fixed time length of the vibration signal is uploaded to the cloud data center according to the set period.

[0022] The application also provides an elevator abnormal state online monitoring system based on vibration signal analysis, comprising:

[0023] The data acquisition module is used for collecting the original vibration signal of the elevator.

[0024] The central processing module is used for replacing the convolution layer in the DRSN-CW network with a deep separable convolution layer based on the DRSN-CW network, obtaining the DS-DRSN-CW network, and using the trained DS-DRSN-CW network to perform feature extraction and fault identification on the original vibration signal of the elevator.

[0025] The wireless communication module is used for uploading the abnormal vibration signal and the fault identification result to the cloud data center through wireless communication when the DS-DRSN-CW network identifies the abnormal vibration signal.

[0026] Specifically, the data acquisition module comprises four groups of MEMS acceleration sensors arranged on the elevator traction machine, the left guide rail of the elevator, the right guide rail of the elevator and the elevator car respectively.

[0027] Preferably, the elevator abnormal state online monitoring system further comprises the following modules:

[0028] The clock synchronization module is used for accurately adjusting the local clock frequency and maintaining high-precision time synchronization between the MEMS acceleration sensors.

[0029] Preferably, the central processing module is also used for real-time power consumption optimization of the elevator abnormal state online monitoring system, and specifically comprises:

[0030] The DS-DRSN-CW network performs real-time fault identification on the collected original vibration signal of the elevator, after the wireless communication module is initialized, it is immediately configured to a low-power mode, the data transmission thread enters a waiting state, and the fault identification result flag signal output by the DS-DRSN-CW network is continuously monitored; after the wireless communication module receives the flag signal representing the fault, the low-power mode is closed, the working mode is switched to, the data transmission thread is started, the abnormal vibration signal and the fault identification result are uploaded to the cloud data center, and after the transmission of the abnormal vibration signal and the fault identification result is completed, the wireless communication module is automatically configured to the low-power mode.

[0031] Beneficial effects: Compared with the prior art, the significant effects of the present application are: at the level of original vibration signal processing, in view of the problem of large feature extraction error of traditional methods under complex working conditions, the present application introduces a lightweight deep residual shrinkage network (DRSN), and optimizes and replaces the convolutional layer in the original DRSN network by using a depth separable convolution, through the self-adaptive threshold shrinkage mechanism and the depth separable convolution optimization, while suppressing environmental noise and electromagnetic interference, the model calculation complexity can be effectively reduced, and high-precision extraction of fault features in the elevator vibration signal can be realized. At the level of the perception system, in view of the technical bottleneck of difficult wiring of wired monitoring and limited edge resources of wireless monitoring, the present application constructs a distributed wireless monitoring architecture based on edge computing, realizes real-time collection and wireless transmission of elevator operation data through a low-power sensor node and a multi-mode communication module; at the same time, the edge node embedded with the improved DRSN network completes data preprocessing and preliminary fault identification, and only uploads the key data to the cloud, effectively reducing the bandwidth occupation and transmission delay. At the same time, a dynamic task scheduling mechanism is designed, the wireless communication module is intelligently awakened according to the data characteristics, the system energy consumption is reduced, and the sensor node endurance time is prolonged. Through the dual breakthroughs of algorithm optimization and system architecture innovation, the present application realizes wireless real-time perception of elevator fault signals, while ensuring detection accuracy, significantly improves the lightness and reliability of the system, and provides an efficient solution for elevator intelligent operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is a method flowchart of the present application.

[0033] Figure 2 is a structural schematic diagram of the RSBU-CW module.

[0034] Figure 3 is a structural schematic diagram of the DS-RSBU-CW module of the present application.

[0035] Figure 4 is a structural schematic diagram of the DS-DRSN-CW network of the present application.

[0036] Figure 5 is a structural schematic diagram of the system of the present application. DETAILED DESCRIPTION

[0037] The preferred schemes of the present application will be further described below in combination with the drawings.

[0038] Example 1

[0039] Please refer to Figure 1 , the present embodiment provides an elevator abnormal state online monitoring method based on vibration signal analysis, including the following steps:

[0040] S1, collecting the original vibration signal of the elevator by using the vibration signal sensor arranged on different components of the elevator.

[0041] S2, to solve the noise interference problem of the original vibration signal of the elevator under complex working conditions, while meeting the calculation resource limitation of the edge device, the DRSN-CW network model is selected, the core of which is to independently set a threshold value for each feature channel to realize more refined noise suppression and feature extraction; please refer to Figure 2 The core of the DRSN-CW network is the RSBU-CW module (Residual Shrinkage Building Unit-CW), the input of the RSBU-CW module is first subjected to two convolution (Conv) operations, then the channel statistical information is obtained through global average pooling (GAP), and then the channel threshold value is dynamically generated through two fully connected networks (FC) and Sigmoid activation function, and the channel threshold value is used to design a soft threshold function for screening.

[0042] The soft threshold function formula is:

[0043]

[0044] In the formula, y is the output feature, x is the input feature, and τ is the channel threshold value.

[0045] The channel threshold value is not a fixed value set, but is generated adaptively through the network structure, specifically, the channel threshold value is obtained through the following steps:

[0046] The global average pooling is used to make global average of the feature matrix of the channel (matrix dimension: H*W) to obtain a scalar representing the overall feature intensity of the channel, and the scalar formula is:

[0047]

[0048] In the formula, S C is the scalar of channel C, X is the channel feature matrix, H is the height, and W is the width.

[0049] The scalar is subjected to nonlinear transformation and dimension adjustment through the fully connected network, and then the result is mapped to the interval (0, 1) through the Sigmoid activation function, and the channel threshold value corresponding to the channel is output in combination with the amplitude range of the channel feature (statistical extreme value and mean value of the feature absolute value).

[0050] In view of the demand of edge side deployment scene for model lightweight, the application replaces the convolution layer Conv in the DRSN-CW network with a depth separable convolution layer DSC on the basis of the DRSN-CW network. The depth separable convolution divides the convolution operation into two independent stages of depth convolution and point-by-point convolution, which significantly reduces the computational complexity and parameter size. The depth convolution is responsible for local feature extraction in the channel dimension, and the point-by-point convolution realizes cross-channel feature interaction and fusion through one-dimensional convolution operation. This decomposition method not only effectively reduces the number of model parameters, but also greatly reduces the computational overhead while maintaining the expression ability of the network.

[0051] Please refer to Figure 3 By replacing the convolution layer in the RSBU-CW module with a depth separable convolution layer, a DS-RSBU-CW (Depthwise Separable-RSBU-CW) module is obtained, and a DS-DRSN-CW (Depthwise Separable-DRSN-CW) network as shown in Figure 4 The DS-DRSN-CW network includes an input layer, a depth separable convolution layer, a plurality of DS-RSBU-CW layers, a global average pooling layer and a fully connected layer connected in sequence.

[0052] The specific steps of the DS-DRSN-CW network model training process are as follows:

[0053] (1) Data segmentation: The original vibration signal is segmented by 2048 sampling points per segment to ensure complete capture of elevator component vibration characteristics.

[0054] (2) Feature extraction: The vibration features are effectively extracted by wavelet decomposition and wavelet packet decomposition, and then the original data, wavelet decomposition results and wavelet packet decomposition results are combined to form a multi-modal signal with a format of 3x2048, which is used as the model input to improve the model's ability to extract vibration features.

[0055] (3) Data standardization: Z-score standardization is used to normalize the three modal data based on the mean and standard deviation of the training set.

[0056] (4) Data set division: The data set is divided into training set, validation set and test set in the ratio of 7:2:1 to ensure that the same elevator data does not cross the set.

[0057] (5) Model training: The Adam optimizer is selected, the initial learning rate is set to 0.001, the cross-entropy is used as the loss function, the batch size is set to 32 to balance the memory limit of the edge device and the data feature capture, the total training rounds are set to 30, and the early stopping mechanism is combined to prevent overfitting.

[0058] (6) Model quantization: after training, the model is quantized to INT8 precision with TensorFlow Lite, strictly controlling the edge inference accuracy loss within 3%, ensuring efficient operation of the model on edge devices.

[0059] The trained DS-DRSN-CW network is used to extract features and identify faults from the original elevator vibration signals.

[0060] S3, when the DS-DRSN-CW network identifies abnormal vibration signals, upload the abnormal vibration signals and fault identification results to the cloud data center through wireless communication. Specifically, the fault identification results output by the DS-DRSN-CW network are divided into normal and abnormal states, corresponding to normal and abnormal vibration signals. If the output is abnormal, upload the abnormal vibration signals and fault identification results to the cloud data center through wireless communication; if the output is normal, upload the vibration signals of fixed length to the cloud data center according to the set period for state declaration.

[0061] The cloud data center receives data uploaded by each edge computing node set in different elevators, and judges the rationality of the fault and issues a warning through the following process: correlation analysis is performed on the data of multiple elevators in the same area, for example, if similar abnormal vibrations occur in adjacent elevators within a similar time period, it may indicate a regional problem (such as power supply anomaly); if the data of a single elevator shows an abnormality, combined with historical operation data and equipment maintenance records, determine whether the abnormal vibration signal corresponds to a real fault. Finally, save the processed data to the cloud database, and according to the severity of the fault, send fault warning information to the maintenance personnel through SMS, email or a dedicated management platform.

[0062] Embodiment 2

[0063] Please refer to Figure 5 The embodiment provides an elevator abnormal state online monitoring system based on vibration signal analysis, which includes a central processing module, a data acquisition module, a wireless communication module, a clock synchronization module and an energy management module:

[0064] Data acquisition module: including but not limited to multiple groups of MEMS acceleration sensors respectively arranged on the elevator traction machine, the left guide rail of the elevator, the right guide rail of the elevator and the elevator car, which can achieve a sampling frequency of up to 10kHz, support synchronous acquisition of three-axis vibration signals, and directly output digital signals to ensure the integrity and accuracy of the original vibration signal data.

[0065] Central processing module: taking STM32 microcontroller (MCU, including CPU + RAM + ROM + peripherals) as the core, it undertakes the central control function of the entire hardware system, can process the original elevator vibration signals transmitted by the data acquisition module through the DS-DRSN-CW network designed in embodiment 1, and perform tasks including but not limited to noise suppression, feature extraction and fault diagnosis; at the same time, it can dynamically adjust the working mode and sampling frequency of each module according to the battery power state feedback by the energy management module, reduce power consumption on the premise of ensuring data integrity; according to the calibration signal of the clock synchronization module, coordinate the multi-node data acquisition timing; control the sleep and wake-up cycle of the wireless communication module, activate the communication link only when fault data is detected or control instructions are received, and ensure stable and efficient operation of the system.

[0066] Wireless communication module: designed in Bluetooth and 4G dual mode, the Bluetooth module is mainly responsible for short-distance and low-power communication between the central processing module and the field control center, used for device debugging, parameter configuration and other scenes; the 4G module undertakes the long-distance data transmission task between the central processing module and the cloud data center, ensures real-time uploading of fault data. When the DS-DRSN-CW network identifies abnormal vibration signals, it uploads the abnormal vibration signals and fault identification results to the cloud data center through the wireless communication module.

[0067] The energy management module is powered by a lithium battery, integrates power monitoring and intelligent charging and discharging management circuit, monitors battery capacity, voltage and other parameters in real time, and dynamically adjusts the power supply strategy according to the system running state, prolongs the service life of the battery.

[0068] Clock synchronization module: based on voltage-controlled temperature-compensated crystal oscillator, the local clock frequency is accurately adjusted through clock taming, realizing high-precision time synchronization sampling control between MEMS acceleration sensors, with synchronization error controlled within ±50μs, ensuring the time consistency of multi-source vibration data, laying a foundation for subsequent signal analysis and fault positioning.

[0069] The task scheduling process of the central processing module control elevator abnormal state online monitoring system includes:

[0070] Data discrimination and module initialization: before data transmission, the DS-DRSN-CW network of the central processing module deployed on the edge platform discriminates the collected original vibration signals in real time; after the wireless communication module is initialized, it is immediately configured in low-power mode, and the data transmission thread enters the waiting state, continuously monitoring the flag signal sent by the data judgment thread.

[0071] Abnormal data transmission: When the data judgment thread determines that there is faulty data based on the DS-DRSN-CW network judgment, a flag bit is sent to the data transmission thread. After receiving the flag bit, the wireless communication module closes the low-power mode, switches to the working mode, and starts the data transmission thread. The transmission content includes original vibration signals, fault diagnosis results, fault occurrence time, and other information, ensuring that the cloud data center can fully obtain abnormal situations.

[0072] Low-power waiting cycle: After completing the data transmission task, the wireless communication module automatically returns to the low-power mode and waits for the next round of faulty data. At the same time, the system regularly performs self-checking to check the running state of each module and the battery power. If the power is lower than the set threshold, the sampling frequency and communication strategy are adjusted to further reduce energy consumption and ensure the continuous and stable operation of the system.

[0073] The application of the embodiment in specific scenarios is described below.

[0074] Residential elevator routine operation monitoring: In the residential community scenario, elevators are frequently used and have a large number of people, and abnormal conditions may cause safety accidents such as people being trapped. After deploying the elevator abnormal state online monitoring system, the focus is on monitoring the abnormal vibration of the car and the door machine during the opening and closing process. When detecting that the vibration amplitude of the car abnormally increases or the vibration frequency abnormally fluctuates during the opening and closing of the door machine, whether it is a real fault is determined based on historical data, and maintenance personnel are notified in a timely manner to handle the problem and prevent the fault from expanding.

[0075] Commercial complex high-speed elevator monitoring: High-speed elevators in commercial complexes run at high speed and carry heavy loads, requiring higher real-time and accuracy of the system. After deploying the elevator abnormal state online monitoring system, by focusing on monitoring the vibration signals of key components such as the traction machine and brake, problems such as component wear and brake abnormalities can be detected in advance. For example, when periodic impact characteristics appear in the brake vibration signal, it can be determined whether the brake pad is excessively worn based on temperature sensor data, and maintenance can be arranged in a timely manner to ensure the safe and stable operation of the elevator.

[0076] Factory freight elevator overload monitoring: Factory freight elevators often carry heavy loads, with large changes in load, which can easily cause mechanical components to overload and deform. After deploying the elevator abnormal state online monitoring system, by analyzing vibration signals and load data collected by sensors, it can be detected whether there are overloads, unbalanced loads, and other situations. Once an anomaly is detected, an alarm is immediately issued, and the elevator is prohibited from running to prevent safety accidents caused by overloading and reduce the risk of equipment damage.

Claims

1. A method for online monitoring of elevator abnormal states based on vibration signal analysis, characterized in that, Includes the following steps: S1. Collect the original vibration signal of the elevator; S2. Based on the DRSN-CW network, replace the convolutional layers in the DRSN-CW network with depth-separable convolutional layers to obtain the DS-DRSN-CW network. Use the trained DS-DRSN-CW network to extract features and identify faults in the original vibration signal of the elevator. S3. When the DS-DRSN-CW network detects an abnormal vibration signal, it uploads the abnormal vibration signal and fault identification results to the cloud data center via wireless communication.

2. The online monitoring method for abnormal elevator conditions according to claim 1, characterized in that: The DS-DRSN-CW network includes a sequentially connected input layer, a depthwise separable convolutional layer, several DS-RSBU-CW layers, a global average pooling layer, and a fully connected layer. The DS-RSBU-CW layers are constructed by replacing the convolutional layers of the RSBU-CW modules in the DRSN-CW network with depthwise separable convolutional layers.

3. The online monitoring method for abnormal elevator conditions according to claim 2, characterized in that: The DS-RSBU-CW layer obtains channel statistics through global average pooling, then dynamically generates channel thresholds through two fully connected layers and a Sigmoid activation function, and uses the channel thresholds to design a soft threshold function for filtering.

4. The online monitoring method for abnormal elevator conditions according to claim 3, characterized in that: The formula for the soft threshold function is: In the formula: y is the output feature, x is the input feature, and τ is the channel threshold.

5. The online monitoring method for elevator abnormal conditions according to claim 4, characterized in that: The channel threshold is obtained through the following steps: Global average pooling is used to perform a global average on the feature matrix of a channel, resulting in a scalar representing the overall feature intensity of that channel. The scalar formula is as follows: In the formula: S C Let X be the scalar of channel C, X be the channel feature matrix, H be the height, and W be the width. The scalar is nonlinearly transformed and its dimensions are adjusted through a fully connected network. Then, the result is mapped to the (0,1) interval through the Sigmoid activation function. The amplitude range of the channel features is combined to output the channel threshold corresponding to that channel.

6. The online monitoring method for abnormal elevator conditions according to claim 1, characterized in that: Step S3 includes: The fault identification results output by the DS-DRSN-CW network are divided into two states: normal and abnormal. If the output is abnormal, the abnormal vibration signal and fault identification results are uploaded to the cloud data center via wireless communication. If the output is normal, the vibration signal of a fixed duration is uploaded to the cloud data center according to the set period.

7. An online monitoring system for abnormal elevator conditions based on vibration signal analysis, characterized in that, include: Data acquisition module: used to acquire the elevator's original vibration signals; Central processing module: Based on the DRSN-CW network, it replaces the convolutional layers in the DRSN-CW network with depthwise separable convolutional layers to obtain the DS-DRSN-CW network, and uses the trained DS-DRSN-CW network to extract features and identify faults from the original vibration signal of the elevator. Wireless communication module: When the DS-DRSN-CW network detects an abnormal vibration signal, it uploads the abnormal vibration signal and fault identification results to the cloud data center via wireless communication.

8. The elevator abnormality online monitoring system according to claim 7, characterized in that: The data acquisition module includes four sets of MEMS acceleration sensors respectively installed on the elevator traction machine, the left guide rail of the elevator, the right guide rail of the elevator, and the elevator car.

9. The elevator abnormality online monitoring system according to claim 8, characterized in that, It also includes the following modules: Clock synchronization module: Used to precisely adjust the local clock frequency to maintain high-precision time synchronization between various MEMS accelerometers.

10. The elevator abnormality online monitoring system according to claim 9, characterized in that: The central processing module is also used for real-time power consumption optimization of the elevator abnormal state online monitoring system, specifically including: The DS-DRSN-CW network performs real-time fault identification on the collected raw elevator vibration signals. After the wireless communication module is initialized, it is immediately configured to low-power mode, and the data transmission thread enters a waiting state, continuously monitoring the fault identification result flag signal output by the DS-DRSN-CW network. When the wireless communication module receives the flag signal representing the fault, it turns off the low-power mode, switches to the working mode, starts the data transmission thread, and uploads the abnormal vibration signal and fault identification result to the cloud data center. After the transmission of the abnormal vibration signal and fault identification result is completed, the wireless communication module automatically configures to low-power mode.

Images