A method and system for controlling elevator safety brakes with multi-stage braking

Abstract

This invention proposes a multi-stage braking elevator safety clamp control method and system, belonging to the field of elevator braking technology. It includes a base mounted on the elevator car; an elevator guide rail mounted inside the elevator shaft and extending vertically; two sets of synchronous servo electric telescopic mechanisms, positioned opposite each other and spaced apart on the base, located on opposite sides of the elevator guide rail's extension direction; each set of synchronous servo electric telescopic mechanisms has at least two safety clamps at its output end; an acceleration sensor for acquiring the elevator car's acceleration signal; and an intelligent control unit communicatively connected to the acceleration sensor and the two sets of synchronous servo electric telescopic mechanisms. Based on a dual-branch fault detection algorithm using built-in dynamic convolutional branches and improved LSTM branches, when a fault is detected in the elevator, the control unit activates the two sets of synchronous servo electric telescopic mechanisms to perform staged braking of the elevator car when the elevator's acceleration exceeds a set threshold.

Description

Technical Field

[0001] This invention relates to the field of elevator braking technology, and in particular to a multi-stage braking elevator safety clamp control method and system. Background Technology

[0002] During elevator operation, the elevator safety brake is a device used for emergency braking when the elevator malfunctions. The safety brake and the speed governor are linked. When the elevator overspeeds, the speed governor triggers the safety brake. The working principle of the mainstream safety brakes on the market is as follows: when the elevator speed exceeds 1.15 times the rated speed, the speed governor activates, pulling the lifting device of the safety brake via a steel cable, causing the wedges to clamp the elevator guide rails and brake the car. The structure of the mainstream safety brake includes a lifting mechanism and a braking mechanism. The lifting mechanism transmits the action of the speed governor, and the wedges in the braking mechanism contact the elevator guide rails to generate friction. This mainstream safety brake relies on the linkage with the speed governor, and the linkage device is prone to failure, failing to trigger braking when overspeeding. It also suffers from insufficient braking force and excessive vibration during braking. Traditional elevator safety brakes have the following main disadvantages: 1. High maintenance requirements: The safety brake requires regular inspection and maintenance to ensure it can function properly in emergencies. This includes: 1. Cleaning foreign objects from the jaws, such as sand, dust, and grease, which may obstruct effective contact between the safety clamp wedge and the elevator guide rail; 2. Response time issues: In some cases, the lifting mechanism of the safety clamp may fail to respond in time due to incorrect dimensions or inadequate lubrication, resulting in insufficient lifting stroke and preventing the wedge from contacting the working surface of the elevator guide rail, causing failure; 3. Structural dimension issues: The safety clamp design may have dimensional mismatch issues, such as insufficient lifting rod stroke, causing the wedge to fail to clamp the elevator guide rail in time when the safety clamp is triggered; 4. Environmental impact: Harsh elevator operating environments, such as dust and oil, may affect the performance of the safety clamp, leading to inaccurate braking or failure; 5. Technical limitations: Traditional safety clamps may not be suitable for all types of elevators or all operating conditions. For example, in high-speed elevators, traditional safety clamps may not provide sufficient braking force.

[0003] Therefore, it is essential to provide a multi-stage braking elevator safety gear control method and system to improve the existing safety gear's reliance on speed governor linkage, the susceptibility of linkage device failure leading to untimely braking, large vibration during braking, and insufficient braking force resulting in unreliable braking. Summary of the Invention

[0004] In view of this, the present invention proposes an elevator safety clamp control method and system that clamps the elevator guide rail at different positions, increases braking force and reduces braking noise, analyzes and processes acceleration sensor signals, and avoids the impact of abnormal elevator vibration on the multi-stage braking of the elevator safety clamp.

[0005] On one hand, the present invention provides a multi-stage braking elevator safety gear control method, comprising:

[0006] Two sets of synchronous servo electric telescopic mechanisms are symmetrically arranged on both sides of the elevator guide rail; the two sets of synchronous servo electric telescopic mechanisms are arranged opposite each other and spaced apart on the base at the bottom of the elevator car, and the output ends of the two sets of synchronous servo electric telescopic mechanisms are respectively equipped with at least two safety clamps.

[0007] Configure an acceleration sensor; the acceleration sensor is installed on the elevator car to acquire and output the acceleration sensor data sequence of the elevator car operation;

[0008] The system is equipped with an intelligent control unit. The intelligent control unit is connected to the acceleration sensor and two sets of synchronous servo electric telescopic mechanisms to acquire the acceleration sensor data sequence. Based on the dual-branch fault detection algorithm with built-in dynamic convolution branch and improved LSTM branch, when the elevator is found to have a fault, the system activates the two sets of synchronous servo electric telescopic mechanisms so that the safety clamps abut against the surfaces of different positions on the elevator guide rails to perform graded braking on the elevator car.

[0009] Based on the above technical solutions, the preferred dual-branch fault detection algorithm, which combines dynamic convolutional branches and improved LSTM branches, includes the following:

[0010] 1) Data preprocessing;

[0011] 2) Dynamic convolution branch: A set of filters with multiple different convolution kernels is preset. By introducing dynamically calculated attention weights, the filters are dynamically combined to generate the optimal convolution kernel for the current input. The dynamic convolution branch is used to output the first fault prediction result.

[0012] 3) Improved LSTM branch: Construct a time series anomaly detection model based on a Long Short-Term Memory (LSTM) network and an encoder-decoder structure. Both the encoder and decoder include multiple stacked LSTM networks. The accelerometer data sequence is fed into the encoder as input. The encoder learns the temporal dependencies in the input accelerometer data sequence, captures the temporal features in the accelerometer data sequence and weights them. The weighted accelerometer data sequence is fed into the decoder. The decoder extracts the hidden information and constructs the output sequence. The reconstruction loss of the input accelerometer data sequence and the output sequence is obtained. The improved LSTM branch outputs a second fault prediction result based on the reconstruction loss.

[0013] 4) The first fault prediction result output by the dynamic convolution branch and the second fault prediction result output by the improved LSTM branch are fused to obtain the output result of whether the elevator car acceleration signal is abnormal.

[0014] Preferably, the data preprocessing involves performing the following perturbation processing on the original accelerometer data sequence during dataset creation to generate multiple pseudo-samples to expand the training set: shifting the time axis of the original accelerometer data sequence by 0.2-1.0 seconds to simulate subsequent sampling or sampling drift; adding Gaussian noise to local segments of the original accelerometer data sequence to simulate electromagnetic interference; reversing short-time sequences as "non-fault interference" samples; artificially masking some abnormal peaks; and performing sliding slicing processing on the original accelerometer data sequence using time windows of different widths to ensure that the same type of noise or anomaly has training samples at different time scales.

[0015] Preferably, the process of obtaining the reconstruction loss of the input accelerometer data sequence and the output sequence involves constructing a maximum time-weighted reconstruction loss model, amplifying the weights of severe anomalies, calculating the reconstruction loss of each point in the data sequence, and then converting the reconstruction loss of each point into a probability value.

[0016] Preferably, the step of fusing the first fault prediction result output by the dynamic convolution branch and the second fault prediction result output by the improved LSTM branch to obtain the output result of whether the elevator car acceleration signal is abnormal involves setting a fault determination threshold and outputting an elevator fault or normal result based on the relationship between the fused value and the fault determination threshold.

[0017] Based on the above technical solution, preferably, the two sets of synchronous servo electric telescopic mechanisms further include at least two servo motors and at least two electric cylinders respectively; the at least two servo motors and at least two electric cylinders are connected in a one-to-one correspondence; the at least two electric cylinders are spaced apart along the extension direction of the elevator guide rail, and each of the at least two electric cylinders includes a fixed end and a movable end, the fixed end is fixedly connected to the base, the movable end extends toward the elevator guide rail, and a safety clamp is provided at the end of the movable end away from the fixed end, and the movable end is hinged to the safety clamp; the safety clamps of the two sets of synchronous servo electric telescopic mechanisms at the same height operate simultaneously.

[0018] Preferably, the step of performing graded braking on the elevator car specifically involves the intelligent control unit acquiring the acceleration signal collected by the acceleration sensor, and then using a dual-branch fault detection algorithm consisting of a built-in dynamic convolution branch and an improved LSTM branch to detect whether the acceleration signal is abnormal, and continuously outputting the result of whether the elevator car acceleration is abnormal; if the elevator car acceleration is greater than a set threshold, it is determined that the elevator car acceleration is abnormal, and the intelligent control unit drives two sets of synchronous servo electric telescopic mechanisms to simultaneously activate the uppermost pair of safety clamps, maintaining the clamping of one end of the guide rail for a period of time. If the elevator car acceleration is not 0, the intelligent control unit drives two sets of synchronous servo electric telescopic mechanisms to simultaneously activate the other pair of safety clamps adjacent to the uppermost position, further clamping the guide rail, and so on, until all safety clamps clamp the elevator guide rail, causing the elevator car to completely stop running.

[0019] Preferably, when the safety clamps of the two sets of synchronous servo electric telescopic mechanisms at the same height operate simultaneously, the friction force against the elevator guide rail increases linearly and reaches the maximum friction force.

[0020] Preferably, the set threshold for elevator car acceleration is 3 m / s². 2 .

[0021] On the other hand, the present invention provides a multi-stage braking elevator safety gear control system for implementing the above-mentioned multi-stage braking elevator safety gear control method, comprising:

[0022] Two sets of synchronous servo electric telescopic mechanisms are installed on the elevator car and are symmetrically arranged on both sides of the elevator guide rail extension direction, with a gap between them and the elevator guide rail. At least two safety clamps are provided on the output end of the two sets of synchronous servo electric telescopic mechanisms on the side closest to the elevator guide rail.

[0023] An accelerometer, installed on the elevator car, is used to acquire and output a sequence of accelerometer data related to the elevator car's movement.

[0024] The intelligent control unit is communicatively connected to the accelerometer and two sets of synchronous servo electric telescopic mechanisms to acquire accelerometer data sequences. The intelligent control unit incorporates a dual-branch fault detection algorithm with dynamic convolution and an improved LSTM branch to determine if the elevator is faulty. When a fault is detected, the intelligent control unit activates the two sets of synchronous servo electric telescopic mechanisms, causing the safety clamps to abut against different surfaces of the elevator guide rails, thus applying staged braking to the elevator car.

[0025] The present invention provides a multi-stage braking elevator safety gear control method and system, which has the following advantages compared with the prior art:

[0026] (1) By configuring multi-stage braking safety clamp structures on both sides of the elevator guide rail, the safety clamps at the same position will be linked together. In the event of abnormal speed or acceleration of the elevator, one or more pairs of safety clamps of the two sets of synchronous servo electric telescopic mechanisms will be activated immediately to ensure that the elevator can be reliably stopped in the shortest time.

[0027] (2) During elevator operation, mechanical vibration, electromagnetic interference, etc. may cause short-term abnormal fluctuations in sensor data, elevator door impact, short-term impact of passengers carrying goods may cause sudden changes in acceleration or speed signals. The dual-branch fault detection algorithm of dynamic convolution branch and improved LSTM branch is used to detect whether the elevator operation status is abnormal, so as to start the safety gear in time and accurately when the elevator malfunctions, and avoid the abnormal interference signal causing the safety gear to start incorrectly. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the equipment assembly position for a multi-stage braking elevator safety clamp control method and system according to the present invention.

[0030] Figure 2 This is a schematic diagram showing the positions of two sets of synchronous servo electric telescopic mechanisms relative to the elevator guide rails in the multi-stage braking elevator safety clamp control method and system of the present invention.

[0031] Figure 3 This is a perspective view of a synchronous servo electric telescopic mechanism of a multi-stage braking elevator safety clamp control method and system according to the present invention.

[0032] Figure 4 This is a schematic diagram of the hinge connection between the electric cylinder and the safety gear of the multi-stage braking elevator safety gear control method and system of the present invention.

[0033] Figure 5 This is a schematic diagram showing the connection between the servo motor and the electric cylinder of the multi-stage braking elevator safety clamp control method and system of the present invention.

[0034] Figure 6 This is a flowchart illustrating the control strategy of a multi-stage braking elevator safety gear control method and system according to the present invention.

[0035] Figure 7This is a structural diagram of the dynamic convolution branch of the multi-stage braking elevator safety gear control method and system of the present invention;

[0036] Figure 8 This invention relates to an improved LSTM branch mechanism for a multi-stage braking elevator safety gear control method and system.

[0037] Figure 9 This is a flowchart of the overall process of a dual-branch fault detection algorithm using dynamic convolutional branches and improved LSTM branches for a multi-stage braking elevator safety gear control method and system according to the present invention.

[0038] Reference numerals: 100, base; 200, elevator guide rail; 300, synchronous servo electric telescopic mechanism; 400, acceleration sensor; 500, intelligent control unit; 1, first-stage safety clamp; 2, first electric cylinder; 3, first servo motor; 4, second servo motor; 5, second electric cylinder; 6, second-stage safety clamp. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0040] Traditional elevator safety brakes have several drawbacks, including high maintenance requirements, foreign objects affecting the contact between the safety brake and the guide rail, slow response, mismatched structural dimensions, susceptibility to environmental influences, and inability to provide sufficient braking force.

[0041] In view of this, such as Figure 1 and Figure 6 As shown, the present invention provides a multi-stage braking elevator safety gear control method, comprising:

[0042] The base (100mm) is installed on the elevator car; such as at the bottom or top of the elevator car. Figure 1 The base 100 is located below the car.

[0043] The elevator guide rail 200 is installed inside the elevator shaft and extends vertically; the elevator car moves linearly along the elevator guide rail 200.

[0044] Two sets of synchronous servo electric telescopic mechanisms 300 are arranged opposite each other and spaced apart on the base 100. The two sets of synchronous servo electric telescopic mechanisms 300 are located on both sides of the extension direction of the elevator guide rail 200. The input ends of the two sets of synchronous servo electric telescopic mechanisms 300 are connected one-to-one with the output ends of the two sets of servo units. The output ends of the two sets of synchronous servo electric telescopic mechanisms 300 are respectively provided with at least two safety clamps. The two sets of synchronous servo electric telescopic mechanisms 300 are arranged symmetrically on both sides of the elevator guide rail 200 to synchronously clamp different positions of the elevator guide rail 200 and realize the braking function.

[0045] An acceleration sensor 400 is used to acquire the acceleration signal of the elevator car during operation. In this embodiment, the acceleration sensor 400 is installed on the elevator car to collect the acceleration signal of the elevator car.

[0046] The intelligent control unit 500 is communicatively connected to the acceleration sensor 400 and two sets of synchronous servo electric telescopic mechanisms 300. It is used to activate the two sets of synchronous servo electric telescopic mechanisms 300 when the elevator's acceleration exceeds a set threshold, causing the safety clamp to abut against the surface of the elevator guide rail 200 and perform graded braking of the elevator car. When the elevator detects that the acceleration exceeds the set threshold or experiences abnormal conditions such as stalling during operation, the intelligent control unit 500 will immediately activate the two sets of synchronous servo electric telescopic mechanisms 300 for emergency braking. The two sets of synchronous servo electric telescopic mechanisms 300 employ a graded braking strategy to ensure that the elevator can stop safely in the shortest possible time.

[0047] like Figure 2 As shown, the two sets of synchronous servo electric telescopic mechanisms 300 also include at least two servo motors and at least two electric cylinders respectively; the at least two servo motors and at least two electric cylinders are connected one-to-one; the at least two electric cylinders are spaced apart along the extension direction of the elevator guide rail 200, and each of the at least two electric cylinders includes a fixed end and a movable end. The fixed end is fixedly connected to the base 100, and the movable end extends toward the elevator guide rail 200. A safety clamp is provided at the end of the movable end away from the fixed end, and the movable end is hinged to the safety clamp; the safety clamps of the two sets of synchronous servo electric telescopic mechanisms 300 at the same height act simultaneously.

[0048] like Figure 2 Combination Figure 3As shown in the diagram, a synchronous servo electric telescopic mechanism 300 with a two-stage braking structure is illustrated. The fixed ends of the first electric cylinder 2 and the second electric cylinder 5 are fixedly connected to different positions on the base 100, such as by bolts, achieving a layered arrangement in height. The first electric cylinder 2 and the second electric cylinder 5 internally employ a multi-stage planetary reducer and a high-precision ball screw composite transmission. Their output ends are connected to the first-stage safety clamp 1 and the second-stage safety clamp 6 via hinges. The first servo motor 3 is connected to the first electric cylinder 2 via a single diaphragm coupling, and the second servo motor 4 is connected to the second electric cylinder 5 via a single diaphragm coupling. Both sides of the synchronous servo electric telescopic mechanism 300 on the elevator guide rail 200 communicate with the intelligent control unit 500 via a CAN bus, ensuring that the braking force deviation of the synchronous servo electric telescopic mechanism 300 on both sides of the elevator guide rail 200 does not exceed 5%. The illustrated two-stage safety clamp synchronous servo electric telescopic mechanism 300 is only an example and is not considered a limitation on the number of safety clamps in the solution. In actual use, more electric cylinders, servo motors, and safety clamps can be set as needed.

[0049] The braking process in this embodiment is as follows:

[0050] During emergency braking, the first-stage safety brake 1, driven by the first electric cylinder 2, is activated. The first-stage safety brake 1 quickly clamps the elevator guide rail 200, initiating deceleration of the elevator car. Simultaneously, the acceleration sensor 400 continues to monitor the elevator's operating status to confirm the effectiveness of the braking. If the braking effect of the first-stage safety brake 1 is insufficient, or if the elevator acceleration has not yet decreased to a safe range, the second-stage braking will automatically commence. The second electric cylinder 5 is activated, driving the second-stage safety brake 6, which gradually clamps the elevator guide rail 200, providing additional braking force. The two-stage safety brakes work together to ensure the elevator decelerates smoothly and eventually comes to a complete stop. Throughout the braking process, the intelligent control unit 500 continuously receives feedback information from the acceleration sensor to ensure the reliability and safety of the braking system. Through this tiered braking strategy and multiple safety monitoring mechanisms, the safety brakes can respond rapidly in emergencies, effectively protecting passenger safety.

[0051] By combining the intelligent control unit 500 with a multi-stage braking mechanism, stable braking force can be provided for the elevator under different conditions, reducing the safety risks caused by the failure of a single brake. Simultaneously, the multi-stage braking system uses electric cylinders to drive at least two safety clamps to gradually clamp the elevator guide rails, with friction increasing linearly until maximum friction is reached. At this point, the rate of change of acceleration of the elevator car does not reach a large value instantaneously like traditional elevator safety clamps, but rather increases gradually, effectively avoiding the problem of large elevator car vibrations caused by sudden changes in the rate of acceleration, thus preventing further injury to passengers.

[0052] To achieve graded braking of the elevator car, the intelligent control unit 500 acquires the acceleration signal collected by the accelerometer 400. It then uses a dual-branch fault detection algorithm (built-in dynamic convolution and improved LSTM branches) to detect any abnormalities in the acceleration signal and continuously outputs the result indicating whether the elevator car acceleration is abnormal. If the elevator car acceleration exceeds a set threshold, it is determined to be abnormal. The intelligent control unit 500 then drives two sets of synchronous servo electric telescopic mechanisms 300 to simultaneously activate the uppermost pair of safety clamps. After clamping one end of the guide rail for a period of time, if the elevator car acceleration is not zero, the intelligent control unit 500 drives the two sets of synchronous servo electric telescopic mechanisms 300 to simultaneously activate the other pair of safety clamps adjacent to the uppermost position, further clamping the guide rail. This process continues until all safety clamps have clamped the elevator guide rail 200, bringing the elevator car to a complete stop. In this embodiment, the set threshold for elevator car acceleration is 3 m / s². 2 .

[0053] During emergency braking of an elevator, traditional elevator safety brakes rely on mechanical structures, primarily including the brake body, brake blocks, and lifting mechanism. In emergencies, such as elevator overspeeding or suspension rope breakage, the speed governor triggers the lifting mechanism, causing the safety brake to quickly clamp the guide rail and stop the elevator through friction. However, this method suffers from insufficient braking force and significant vibration during braking. Accelerometers are installed on the top of the elevator car to monitor its operating status. However, during elevator operation, mechanical vibrations and electromagnetic interference can cause short-term abnormal fluctuations in sensor data. Furthermore, sudden acceleration changes can occur due to door impacts or short-term impacts from passengers carrying goods, but these do not necessarily indicate a malfunction. These interference signals can affect the activation of multi-stage braking safety brakes, potentially leading to erroneous activation. Simultaneously, the signal characteristics of the elevator differ during start-up, stopping, and constant-speed operation, and traditional fixed-parameter fault detection algorithms cannot adapt to these variations. Traditional methods rely on setting thresholds to judge acceleration changes, but these methods typically cannot dynamically adapt to different fault modes or accurately identify fault signals from multiple factors in real time. Therefore, abnormal acceleration in elevators is often difficult to detect in a timely manner, leading to delays in fault detection and false alarms / missed alarms. To address this, this embodiment proposes a dual-branch fault detection algorithm based on dynamic convolutional branches and an improved LSTM branch. The dynamic convolutional branch extracts abrupt change features, while the improved LSTM branch extracts time-varying patterns, trends, and dynamic dependencies in the data, helping to capture long-term and short-term acceleration variation features. By combining these two deep learning algorithms, fault features can be efficiently extracted from acceleration data, and fault signals and interference signals in acceleration abrupt changes can be distinguished in real time. This provides accurate elevator fault detection, enabling timely activation of the safety brake when an elevator malfunctions and preventing erroneous activation of the safety brake due to abnormal interference signals.

[0054] A two-branch fault detection algorithm using dynamic convolutional branches and improved LSTM branches, including the following:

[0055] 1) Data preprocessing;

[0056] In actual elevator equipment, the sensor data collected initially often lacks manually labeled "fault-normal" comparison tags. Relying on manual screening to create training sets is not only inefficient but also prone to missing "atypical" fault features. Traditional fixed-threshold labeling strategies cannot cope with highly non-stationary signal changes during elevator operation and are prone to mislabeling. Therefore, this embodiment proposes a soft-label + pseudo-label reconstruction mechanism that combines physical features, engineering experience, and statistical anomalies, coupled with a dynamic temporal feature self-enhancement method, to support early pre-training of deep models.

[0057] Data preprocessing involves perturbing the original accelerometer data sequence during dataset creation to generate multiple pseudo-samples to expand the training set: A) Shifting the time axis of the original accelerometer data sequence by 0.2-1.0 seconds to simulate subsequent sampling or sampling drift; B) Adding Gaussian noise to local segments of the original accelerometer data sequence to simulate electromagnetic interference; C) Reversing short-time sequences as "non-fault interference" samples to enhance the ability of the improved LSTM branch to eliminate false faults; D) Artificially masking some abnormal peaks to train the ability to predict anomalies even with missing peaks; E) Using time windows of different widths to perform sliding slicing on the original accelerometer data sequence, such as 1s, 3s, or 5s, to ensure that the same type of noise or anomaly has training samples at different time scales.

[0058] 2) Dynamic Convolution Branch: A set of filters with multiple different convolution kernels is preset. By introducing dynamically calculated attention weights, the filters are dynamically combined to generate the optimal convolution kernel for the current input. The dynamic convolution branch is used to output the first fault prediction result. Y dc The probability value is between [0, 1].

[0059] Dynamic convolution introduces dynamically calculated weights, allowing the convolution kernel to adaptively adjust based on the input data. The formula is expressed as follows: , C k It is the preset number k One convolutional kernel, a k It is the first dynamic calculation k Each attention weight, Traditional fixed convolution kernels struggle to adapt to the dynamic characteristics of elevator sensor signals (such as acceleration, vibration, current, and noise) during multiple stages of elevator operation (starting, constant speed, braking, and door opening / closing). Dynamic convolution, generated through operational state perception and adaptive filtering kernels, can accurately match the non-stationary characteristics of elevator signals.

[0060] In addition, an elevator signal characteristic and dynamic convolution adaptation mechanism are configured: during the start / brake phase, the motor current surges and acceleration changes sharply, requiring a high-frequency filter to capture the transient response; during the constant speed phase, the signal is stable but may contain periodic mechanical vibration noise, requiring a wide kernel to suppress high-frequency interference; during door opening and closing, short-term impact signals (solenoid valve action, door lock collision) require multi-scale kernel joint detection. Therefore, the signal mean / variance is first extracted as contextual features through global average pooling to generate phase-sensitive weights (e.g., during the braking phase, the 3×3 kernel weights are automatically enhanced to capture acceleration abrupt changes).

[0061] Simultaneously, considering the differences in signal characteristics among heterogeneous sensors, a parallel dynamic convolution architecture is designed: Acceleration signals are rich in high-frequency vibration components, so dynamic convolution focuses on activating small-scale kernels (3×3, 5×5); current signals are dominated by low-frequency trends, so dynamic convolution automatically allocates large kernels (7×7, 9×9) to smooth power frequency noise. Therefore, dynamic weights are calculated independently and in parallel for multiple sensor inputs, and cross-modal features are integrated through an early fusion layer. See the attached diagram for the principle of the dynamic convolution branch. Figure 7 .

[0062] 3) Improved LSTM branch: Construct a time series anomaly detection model based on a Long Short-Term Memory (LSTM) network and an encoder-decoder structure. Both the encoder and decoder include multiple stacked LSTM networks. The accelerometer data sequence is fed into the encoder as input. The encoder learns the temporal dependencies in the input accelerometer data sequence, captures the temporal features in the accelerometer data sequence and weights them. The weighted accelerometer data sequence is fed into the decoder. The decoder extracts the hidden information and constructs the output sequence. The reconstruction loss of the input accelerometer data sequence and the output sequence is obtained. The improved LSTM branch outputs a second fault prediction result based on the reconstruction loss.

[0063] Long Short-Term Memory (LSTM) networks are variants of Recurrent Neural Networks (RNNs) specifically designed for processing and predicting time-series data. Particularly in elevator acceleration data, short-term disturbances caused by passengers moving goods or walking within the elevator are common and need to be eliminated to avoid misjudging elevator malfunctions. LSTM networks can distinguish these disturbances from genuine malfunction signals by learning patterns from historical data. LSTM networks can identify periodic fluctuations and trend changes in acceleration data and determine whether the current acceleration exceeds the normal range based on these patterns. For example, the acceleration of an elevator during smooth operation will exhibit a relatively stable time series with occasional slight fluctuations. When acceleration fluctuations that do not conform to this pattern occur, LSTM networks can identify and eliminate these irregular fluctuations. LSTM networks can capture long-term dependencies in elevator acceleration, thereby understanding changes in the elevator's state. For example, a large acceleration abrupt change occurs during an emergency stop, and this abrupt acceleration is significantly different from the fluctuations caused by short-term passenger actions (such as moving items). By learning these different patterns in historical data, the Long Short-Term Memory (LSTM) network can effectively distinguish between normal fluctuations and faults in acceleration.

[0064] The improved LSTM algorithm structure is as follows: Figure 8As shown, this algorithm is a time-series anomaly detection model based on a Long Short-Term Memory (LSTM) network and an encoder-decoder structure. Input time-series data X1, X2, X3, ..., Xt are fed into the encoder. The encoder consists of multiple stacked LSTM layers, which learn and encode the temporal dependencies in the input sequence. Then, before being passed to the decoder, the encoded data passes through an attention mechanism module to capture temporal features in the input data and weight important information. After passing through the attention mechanism module, the weighted data is fed into the decoder. The decoder's structure is similar to the encoder, also containing multiple LSTM layers. The decoder's role is to reconstruct an output sequence that is as close as possible to the input from the implicit information extracted by the encoder. Then, the reconstruction loss is calculated using the input and output sequences to measure the difference between the model's predicted and actual values. To further highlight the anomalousness of "instantaneous reconstruction bursts," this embodiment designs a maximum time-weighted reconstruction loss model to amplify the impact of severe anomalies at high time points. Compared to the average loss model, this model has stronger temporal sensitivity and "anomaly spike capture" capabilities, making it suitable for rapid response triggering in high-speed fault scenarios. The specific formula is as follows:

[0065] Lr This represents the reconstruction losses. X t It is the first input. t One value, It is the reconstruction value predicted by the Long Short-Term Memory (LSTM) network. It involves the squared difference operation. The reconstruction loss is then input into the anomaly probability detection process. By calculating the change in reconstruction loss for each data point and converting it into a probability value, it is determined whether the data point is an anomaly. The formula is as follows: ,in It is the sigmoid function. The formula is as follows: , where is the sigmoid function. Second fault prediction result. Y lstm The probability value is between [0, 1].

[0066] 4) The first fault prediction result from the dynamic convolution branch and the second fault prediction result from the improved LSTM branch are fused to obtain the output result indicating whether the elevator car acceleration signal is abnormal. By combining the outputs of the dynamic convolution branch and the improved LSTM branch, it is possible to more accurately determine whether an elevator fault has occurred. The specific process is as follows:

[0067] (A) Multi-dimensional Information Fusion: Traditional acceleration detection methods are typically based on a single decision rule, while the proposed solution employs a multi-branch structure of dynamic convolution and an improved LSTM to extract data features from different perspectives: the dynamic convolution branch extracts abrupt change features, while the improved LSTM branch extracts patterns, trends, and dynamic dependencies that change over time in the data, helping to capture both long-term and short-term acceleration variation features. Combining these two different feature branches allows for a more comprehensive assessment of the nature of acceleration changes, thereby improving the accuracy of fault detection.

[0068] (B) Multimodal Information Fusion: The dynamic convolution branch focuses on the local features of abrupt acceleration changes, while the improved LSTM branch focuses on identifying the temporal dependence of acceleration. Fusion of these two types of information enables joint analysis of the multidimensional features of acceleration data, avoiding the limitations of single-feature judgment. Specifically, the dynamic convolution branch handles drastic changes, while the improved LSTM branch eliminates normal fluctuations or interference in the short term, ultimately leading to a judgment of elevator malfunction.

[0069] (C) Output Fusion and Fault Determination: When the dynamic convolution branch detects a sudden change in acceleration, and the improved LSTM branch confirms that the acceleration fluctuation is not an interference signal, an elevator fault is considered to have occurred. This determination mechanism ensures that the fault determination has both sensitivity to local features and long-term consideration of temporal changes by fusing the output results of the dynamic convolution branch and the improved LSTM branch.

[0070] The formula for merging the two branches is as follows: In the denominator It is a non-zero real number. Y This is the final judgment result.

[0071] The final fault determination can be achieved by comparing the fused output with a fault determination threshold. In this embodiment, the given fault determination threshold is 1.5, which satisfies the following condition. If the final judgment result is not less than 1.5, the elevator is deemed to be faulty and requires braking. The two sets of synchronous servo electric telescopic mechanisms 300 adopt a graded braking strategy to stop the elevator. If the final judgment result is less than 1.5, the elevator is considered to be normal and the synchronous servo electric telescopic mechanism 300 is not triggered.

[0072] By combining the outputs of both the dynamic convolutional branch and the improved LSTM branch using the weighted fusion method described above, the likelihood of elevator malfunctions can be assessed more comprehensively. The dynamic convolutional branch is responsible for capturing abrupt changes in the acceleration data, while the improved LSTM branch learns the time-series features of the acceleration data and eliminates interference. The final integrated assessment provides a more accurate elevator malfunction detection mechanism.

[0073] In addition, the present invention provides a multi-stage braking elevator safety gear control system for implementing the above-mentioned multi-stage braking elevator safety gear control method, comprising:

[0074] Two sets of synchronous servo electric telescopic mechanisms are installed on the elevator car and are symmetrically arranged on both sides of the elevator guide rail extension direction, with a gap between them and the elevator guide rail. At least two safety clamps are provided on the output end of the two sets of synchronous servo electric telescopic mechanisms on the side closest to the elevator guide rail.

[0075] An accelerometer, installed on the elevator car, is used to acquire and output a sequence of accelerometer data related to the elevator car's movement.

[0076] The intelligent control unit is communicatively connected to the accelerometer and two sets of synchronous servo electric telescopic mechanisms to acquire accelerometer data sequences. The intelligent control unit incorporates a dual-branch fault detection algorithm with dynamic convolution and an improved LSTM branch to determine if the elevator is faulty. When a fault is detected, the intelligent control unit activates the two sets of synchronous servo electric telescopic mechanisms, causing the safety clamps to abut against different surfaces of the elevator guide rails, thus applying staged braking to the elevator car.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-stage braking elevator safety gear control method, characterized in that, include: Two sets of synchronous servo electric telescopic mechanisms are symmetrically arranged on both sides of the elevator guide rail; the two sets of synchronous servo electric telescopic mechanisms are arranged opposite each other and spaced apart on the base at the bottom of the elevator car, and the output ends of the two sets of synchronous servo electric telescopic mechanisms are respectively equipped with at least two safety clamps. Configure an acceleration sensor; the acceleration sensor is installed on the elevator car to acquire and output the acceleration sensor data sequence of the elevator car operation; The system is equipped with an intelligent control unit. The intelligent control unit is connected to the acceleration sensor and two sets of synchronous servo electric telescopic mechanisms to acquire the acceleration sensor data sequence. Based on the dual-branch fault detection algorithm with built-in dynamic convolution branch and improved LSTM branch, when the elevator is found to have a fault, the system activates the two sets of synchronous servo electric telescopic mechanisms so that the safety clamps abut against the surfaces of different positions on the elevator guide rails to perform graded braking on the elevator car. A two-branch fault detection algorithm using dynamic convolutional branches and improved LSTM branches, including the following: 1) Data preprocessing; 2) Dynamic convolution branch: A set of filters with multiple different convolution kernels is preset. By introducing dynamically calculated attention weights, the filters are dynamically combined to generate the optimal convolution kernel for the current input. The dynamic convolution branch is used to output the first fault prediction result. 3) Improved LSTM branch: Construct a time series anomaly detection model based on a Long Short-Term Memory (LSTM) network and an encoder-decoder structure. Both the encoder and decoder include multiple stacked LSTM networks. The accelerometer data sequence is fed into the encoder as input. The encoder learns the temporal dependencies in the input accelerometer data sequence, captures the temporal features in the accelerometer data sequence and weights them. The weighted accelerometer data sequence is fed into the decoder. The decoder extracts the hidden information and constructs the output sequence. The reconstruction loss of the input accelerometer data sequence and the output sequence is obtained. The improved LSTM branch outputs a second fault prediction result based on the reconstruction loss. 4) The first fault prediction result output by the dynamic convolution branch and the second fault prediction result output by the improved LSTM branch are fused to obtain the output result of whether the elevator car acceleration signal is abnormal.

2. The elevator safety gear control method with multi-stage braking according to claim 1, characterized in that, The data preprocessing involves perturbing the original accelerometer data sequence during dataset creation to generate multiple pseudo-samples to expand the training set: the time axis of the original accelerometer data sequence is shifted by 0.2-1.0 seconds to simulate subsequent sampling or sampling drift; Gaussian noise is added to local segments of the original accelerometer data sequence to simulate electromagnetic interference; short-time series are inverted as "non-fault interference" samples; some abnormal peaks are artificially masked; and sliding slice processing of the original accelerometer data sequence is performed using time windows of different widths to ensure that training samples of the same type of noise or anomaly are available at different time scales.

3. The elevator safety gear control method with multi-stage braking according to claim 1, characterized in that, The reconstruction loss of the input accelerometer data sequence and the output sequence is obtained by constructing a maximum time-weighted reconstruction loss model, amplifying the weight of severe anomalies, calculating the reconstruction loss of each point in the data sequence, and then converting the reconstruction loss of each point into a probability value.

4. The elevator safety gear control method with multi-stage braking according to claim 1, characterized in that, The process of fusing the first fault prediction result output by the dynamic convolution branch and the second fault prediction result output by the improved LSTM branch to obtain the output result of whether the elevator car acceleration signal is abnormal involves setting a fault determination threshold and outputting an elevator fault or normal result based on the relationship between the fused value and the fault determination threshold.

5. The elevator safety gear control method with multi-stage braking according to claim 1, characterized in that, The two sets of synchronous servo electric telescopic mechanisms also include at least two servo motors and at least two electric cylinders respectively; the at least two servo motors and at least two electric cylinders are connected in a one-to-one correspondence; the at least two electric cylinders are spaced apart along the extension direction of the elevator guide rail, and each of the at least two electric cylinders includes a fixed end and a movable end. The fixed end is fixedly connected to the base, and the movable end extends toward the elevator guide rail. A safety clamp is provided at the end of the movable end away from the fixed end, and the movable end is hinged to the safety clamp; the safety clamps of the two sets of synchronous servo electric telescopic mechanisms at the same height operate simultaneously.

6. The elevator safety gear control method with multi-stage braking according to claim 1, characterized in that, The step-by-step braking of the elevator car involves the following steps: After acquiring the acceleration signal from the acceleration sensor, the intelligent control unit uses a dual-branch fault detection algorithm (built-in dynamic convolution and improved LSTM branches) to detect whether the acceleration signal is abnormal and continuously outputs the result indicating whether the elevator car acceleration is abnormal. If the elevator car acceleration exceeds a set threshold, it is determined that the elevator car acceleration is abnormal. The intelligent control unit then drives two sets of synchronous servo electric telescopic mechanisms to simultaneously activate the uppermost pair of safety clamps. After maintaining the clamping of one end of the guide rail for a certain period, if the elevator car acceleration is not zero, the intelligent control unit drives the two sets of synchronous servo electric telescopic mechanisms to simultaneously activate the other pair of safety clamps adjacent to the uppermost position, further clamping the guide rail. This process continues until all safety clamps have clamped the elevator guide rail, causing the elevator car to come to a complete stop.

7. The elevator safety gear control method with multi-stage braking according to claim 6, characterized in that, When the safety clamps of the two sets of synchronous servo electric telescopic mechanisms at the same height operate simultaneously, the friction force against the elevator guide rail increases linearly and reaches the maximum friction force.

8. The elevator safety gear control method with multi-stage braking according to claim 6, characterized in that, The set threshold for elevator car acceleration is 3 m / s². 2 .

9. A multi-stage braking elevator safety gear control system, used to implement the multi-stage braking elevator safety gear control method according to any one of claims 1-8, characterized in that, include: Two sets of synchronous servo electric telescopic mechanisms are installed on the elevator car and are symmetrically arranged on both sides of the elevator guide rail extension direction, with a gap between them and the elevator guide rail. At least two safety clamps are provided on the output end of the two sets of synchronous servo electric telescopic mechanisms on the side closest to the elevator guide rail. An accelerometer, installed on the elevator car, is used to acquire and output a sequence of accelerometer data related to the elevator car's movement. The intelligent control unit is communicatively connected to the accelerometer and two sets of synchronous servo electric telescopic mechanisms to acquire accelerometer data sequences. The intelligent control unit incorporates a dual-branch fault detection algorithm with dynamic convolution and an improved LSTM branch to determine if the elevator is faulty. When a fault is detected, the intelligent control unit activates the two sets of synchronous servo electric telescopic mechanisms, causing the safety clamps to abut against different surfaces of the elevator guide rails, thus applying staged braking to the elevator car.

Images