Brake abnormality diagnosis method, electronic device, medium, and program product

Abstract

The application provides a brake abnormality diagnosis method, an electronic device, a medium and a program product. The method comprises the following steps: acquiring a first wheel speed signal, a second wheel speed signal, an actual electric parameter signal, an actual micro switch signal and an actual displacement signal; obtaining a fused wheel speed signal; inputting the fused wheel speed signal into a deep learning model to output a confidence level of each first fault type; determining, according to a fault type with the maximum confidence level, a first standard electric parameter signal, a first standard micro switch signal and a first standard displacement signal corresponding to the fault type with the maximum confidence level in a knowledge graph based on the knowledge graph; determining a first consistency deviation; and when the first consistency deviation is greater than a deviation threshold, determining an abnormality result based on the knowledge graph and a preset strategy of all first fault types. The application improves the accuracy and reliability of brake abnormality diagnosis.

Description

Technical Field

[0001] This invention relates to the field of special equipment technology, and more specifically, to a method for diagnosing brake malfunctions, electronic equipment, media, and program products. Background Technology

[0002] As one of the most important components for ensuring the safe operation of elevators, the brake's failure modes mainly manifest as follows: aging and hydrolysis of the shock-absorbing pads causing jamming; spring breakage, loosening, bias, or directional deviation causing jamming and insufficient braking force; internal jamming of the release wrench leading to brake failure; excessive changes in the brake gap; changes in the friction coefficient of the friction surface; residual magnetism in the coil causing magnetic force attenuation; and excessive temperature rise of the brake. These failures can be summarized into two points: first, whether its braking torque can meet the stopping requirements; and second, whether the brake can correctly and reliably execute the instructions issued by the elevator control cabinet and operate correctly and reliably.

[0003] For diagnosing the aforementioned problems with brakes, in a large number of elevators in use where the standard does not mandate automatic monitoring of braking force, existing methods based on a single sensor or simple thresholds are prone to misjudgment or missed judgment when there is a deviation between the actual operating signal and the typical fault mode. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a brake abnormality diagnosis method, electronic device, medium and program product, which can improve the problem that existing methods based on a single sensor or simple threshold are prone to misjudgment or omission when there is a deviation between the actual operating signal and the typical fault mode.

[0005] To achieve the above technical objectives, the technical solution adopted in this application is as follows:

[0006] In a first aspect, embodiments of this application propose a method for diagnosing brake malfunctions, the method comprising:

[0007] The system acquires a first wheel speed signal, a second wheel speed signal, an actual electrical parameter signal, an actual micro-switch signal, and an actual displacement signal. The first wheel speed signal and the second wheel speed signal are obtained based on a contact sensor and a non-contact sensor, respectively. Both the first wheel speed signal and the second wheel speed signal are time-series data. The first wheel speed signal includes several first acquisition points, and the second wheel speed signal includes several second acquisition points.

[0008] The first wheel speed signal and the second wheel speed signal are input into the fusion model, and the fusion model fuses the first wheel speed signal and the second wheel speed signal to obtain the fused wheel speed signal;

[0009] The fused wheel speed signal is input into a deep learning model, and the confidence level of each first fault type is output through the deep learning model. The number of first fault types is greater than or equal to 2, and all first fault types are stored in the deep learning model.

[0010] Based on the fault type with the highest confidence level, and using a knowledge graph, determine the first standard electrical parameter signal, the first standard micro switch signal, and the first standard displacement signal corresponding to the fault type with the highest confidence level in the knowledge graph.

[0011] Based on the deviations between the actual electrical parameter signal and the first standard electrical parameter signal, the deviations between the actual microswitch signal and the first standard microswitch signal, and the deviations between the actual displacement signal and the first standard displacement signal, a first consistency deviation is determined.

[0012] When the first consistency deviation is greater than the deviation threshold, an abnormal result is determined based on the knowledge graph and the preset strategies for all first fault types.

[0013] According to the first aspect, the determination of abnormal results based on the knowledge graph and the preset strategy for all first fault types includes:

[0014] Based on the knowledge graph, the second fault type corresponding to the second standard electrical parameter signal, the second standard micro switch signal, and the second standard displacement signal is determined, wherein the second consistency deviation determined based on the deviation between the second standard electrical parameter signal and the actual electrical parameter signal, the deviation between the second standard micro switch signal and the actual micro switch signal, and the deviation between the second standard displacement signal and the actual displacement signal is less than the deviation threshold.

[0015] Obtain the confidence level of the first fault type that is the same as the second fault type;

[0016] When the confidence level of the first fault type is greater than or equal to the confidence threshold, the third initial fault type is output as an abnormal result. When the confidence level of the first initial fault type is less than the confidence threshold, both the second fault type and the first initial fault type are output as abnormal results.

[0017] Secondly, embodiments of this application provide an electronic device comprising a processor and a memory coupled together, wherein the memory stores a computer program, and when the computer program is executed by the processor, the electronic device performs the method described in the first aspect.

[0018] Thirdly, embodiments of this application provide a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when run on a computer, causes the computer to perform the method described in the first aspect.

[0019] Fourthly, embodiments of this application propose a program product including a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0020] The invention employing the above technical solution has the following advantages:

[0021] In the technical solution provided in this application, a time-seriesd first wheel speed signal and a second wheel speed signal are acquired by contact sensors and non-contact sensors respectively, while simultaneously acquiring actual electrical parameter signals, actual micro-switch signals, and actual displacement signals. This avoids the limitations of single-sensor monitoring and solves the problem of single-sensor signal distortion in harsh environments. The two wheel speed signals are input into a fusion model to obtain a fused wheel speed signal, improving the stability and accuracy of the wheel speed signal. The fused wheel speed signal is then input into a deep learning model, which outputs at least two first fault types and their corresponding confidence levels. Subsequently, based on the first fault type with the highest confidence level, the corresponding first standard electrical parameter, micro-switch, and displacement signal are matched using a knowledge graph to calculate the first-to-last comparison between the actual signal and the standard signal. This system addresses consistency deviations, enabling the identification of internal hidden faults such as damping pad aging and spring fatigue, as well as the quantitative assessment of brake performance degradation under harsh environments. When the first consistency deviation exceeds a deviation threshold, a second fault type that meets the second consistency deviation standard is selected based on a knowledge graph. By comparing the confidence level of the first fault type corresponding to this second fault type with the confidence threshold, the system identifies and outputs the abnormal result. This solves the problem of misjudgment and omission when there are deviations between actual operating signals and typical fault modes, replacing the subjective judgment of maintenance personnel. It achieves automatic monitoring of the brake, adapting to old elevators and harsh environments such as western plateaus and high humidity areas, alleviating the unreasonable human-machine ratio in elevator maintenance, and improving the accuracy and reliability of brake anomaly diagnosis. Attached Figure Description

[0022] This application can be further illustrated by the non-limiting embodiments given in the accompanying drawings. It should be understood that the following drawings only illustrate some embodiments of this application and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained from these drawings without any inventive effort.

[0023] Figure 1 A flowchart of a brake malfunction diagnosis method provided in an embodiment of this application.

[0024] Figure 2This is a sub-flowchart of S120 provided in an embodiment of this application. Detailed Implementation

[0025] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts are referred to by the same reference numerals in the drawings or description. Implementations not shown or described in the drawings are forms known to those skilled in the art. In the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Please refer to Figure 1 This application provides a brake anomaly diagnosis method, which can be applied to electronic devices and whose steps can be executed or implemented by the electronic device. The electronic device can be, but is not limited to, a personal computer, a smartphone, or other electronic devices. The brake anomaly diagnosis method may include the following steps:

[0027] S110, the first wheel speed signal and the second wheel speed signal are input into the fusion model, and the fusion model fuses the first wheel speed signal and the second wheel speed signal to obtain the fused wheel speed signal;

[0028] S120, the fused wheel speed signal is input into the deep learning model, and the confidence level of each first fault type is output through the deep learning model. The number of the first fault types is greater than or equal to 2, and all the first fault types are stored in the deep learning model.

[0029] S130, based on the fault type with the highest confidence level, determine the first standard electrical parameter signal, the first standard micro switch signal, and the first standard displacement signal in the knowledge graph that correspond to the fault type with the highest confidence level.

[0030] S140, based on the deviation between the actual electrical parameter signal and the first standard electrical parameter signal, the deviation between the actual micro switch signal and the first standard micro switch signal, and the deviation between the actual displacement signal and the first standard displacement signal, a first consistency deviation is determined;

[0031] S150, when the first consistency deviation is greater than the deviation threshold, an abnormal result is determined based on the knowledge graph and the preset strategies for all first fault types.

[0032] Speed ​​parameters are core parameters reflecting the braking performance of the brake and identifying early potential problems. However, in actual elevator operation scenarios, the dynamic operating characteristics of the traction sheave lead to multiple technical bottlenecks in speed parameter measurement. Both ordinary contact and non-contact speed sensing devices have unavoidable defects, namely, the measurement of speed parameters faces problems such as radial vibration and low-frequency vibration of the traction sheave contact surface, which makes it difficult for ordinary contact speed sensing devices to maintain a stable state for a long time, directly affecting the data calculation results; the speed measurement accuracy of non-contact sensors is difficult to meet the requirements.

[0033] This embodiment employs a dual-sensor approach, combining contact and non-contact sensors for collaborative acquisition. This approach balances the continuity of speed signals with potential for high accuracy. The contact sensor provides a first wheel speed signal with a complete time sequence, while the non-contact sensor provides a second wheel speed signal, compensating for the limitations of a single sensor. By filtering noise components from the first wheel speed signal, the impact of vibration interference on the contact sensor data is initially reduced, laying the foundation for subsequent corrections. Furthermore, by extracting anchor points within the stable range, leveraging the characteristic of minimal traction sheave vibration and maximum speed stability during elevator uniform speed operation, a high-precision second acquisition point is selected as the anchor point. This addresses the overall insufficient accuracy of the non-contact sensor, providing a reliable benchmark for contact signal correction. Subsequently, by calculating the vibration drift between the anchor point and the corresponding first acquisition point, the measurement deviation caused by vibration in the contact sensor is quantified. The first wheel speed signal is then corrected based on this vibration drift, reducing signal distortion caused by vibration interference and simultaneously supplementing the missing timestamp of the second wheel speed signal, resolving the issue of incomplete timing in the non-contact signal. Finally, a preset fusion strategy is used to fuse the corrected first and second wheel speed signals.

[0034] The steps of the brake malfunction diagnosis method will be explained in detail below:

[0035] In S110, a contact speed sensor is installed at the end of the traction sheave shaft to collect the instantaneous speed signal of the traction sheave in real time, covering the entire working condition of "uniform speed operation → braking deceleration → brake in place → shutdown steady state". The collected content is "time stamp + contact speed value".

[0036] The non-contact speed sensor is installed on the side of the traction sheave (not the shaft end, with strong vibration resistance) to collect the original speed signal of the traction sheave in real time. The collected content is the same as that of the contact type. It is only used to extract steady-state anchor points and does not participate in continuous speed monitoring.

[0037] Synchronously collect brake electrical parameters (voltage / current), microswitch action sequence (T1, T2), and brake gap displacement data.

[0038] In this embodiment, a high-precision dedicated voltage sensor is used for data acquisition. The LV25-P series voltage sensor (range 0-220V, compatible with elevator brake coil power supply specifications) is selected, with a measurement accuracy ≤ ±0.5%. The sampling frequency is synchronized with the speed sensor (1000Hz) to ensure timing alignment. The sensor is installed inside the brake control cabinet, away from areas with strong electromagnetic interference, and is connected to the brake coil power supply circuit in parallel to avoid signal distortion caused by improper connection. The voltage signal directly reflects the power supply stability of the brake coil and is a core auxiliary parameter for identifying coil faults and power supply anomalies. Under normal operating conditions, the voltage should be stable within ±5% of the rated power supply voltage. Excessive voltage will burn out the coil, while insufficient voltage will lead to insufficient coil attraction and delayed braking response. Voltage fluctuations > ±10% (for 5 consecutive sampling cycles) indicate potential problems such as power supply anomalies, coil aging, or poor contact.

[0039] In this embodiment, the brake electrical parameters (voltage) are collected synchronously with the speed signal. The voltage sensor monitors the instantaneous voltage value at the input of the brake coil in real time, recording one collection point every 1ms to generate time-series structured data of "timestamp + voltage value". During the collection process, the 3σ criterion is used to automatically remove instantaneous pulse data (values ​​exceeding the mean ± 3 times the standard deviation) caused by electromagnetic interference, retaining the valid voltage signal. At the same time, the energization and de-energization times of the coil are recorded synchronously and bound to the timing of the speed signal and the micro switch signal to ensure the timing consistency during subsequent collaborative diagnosis. The collected valid data is classified and archived according to the braking cycle and stored in the local cache in real time, providing a quantitative basis for subsequent expert database matching and fault verification.

[0040] This embodiment can use the AZ-7141 series high-precision travel microswitches (with a dedicated signal acquisition module), where T1 corresponds to the left brake shoe microswitch and T2 corresponds to the right brake shoe microswitch. The signal acquisition module has a sampling accuracy of ≤±0.1ms and shares a unified system timestamp reference with the speed sensor and electrical parameter sensor to ensure timing synchronization. The microswitches are respectively installed on the brake arms of the left and right brake shoes of the brake. The switch contacts are linked with the brake arms, and the switch can be precisely triggered to open or close when the brake shoes move. During installation, the contact position needs to be adjusted to ensure that the contacts are just pressed when the brake shoes are fully braked, avoiding untimely or false triggering.

[0041] In this embodiment, T1 represents the moment when the left brake shoe reaches the braking position, and T2 represents the moment when the right brake shoe reaches the braking position. Both reflect the synchronicity of the brake shoe's action and the braking response speed, and are key parameters for identifying potential problems such as mechanical jamming of the brake shoes, abnormal microswitch operation, and asynchronous braking. Under normal operating conditions, the response delay between T1, T2, and the braking command should be ≤0.005s, and the time difference between T1 and T2 should be ≤0.002s. If the response delay is too long, the time difference between the left and right switches is too large, or the switches cannot be triggered or cannot be reset after triggering, these all indicate potential problems such as aging microswitch contacts, mechanical jamming of the brake shoes, and uneven braking gap, which will directly affect the braking effect.

[0042] In this embodiment, the signal acquisition module monitors the on / off status of two microswitches in real time. After the braking command is issued, the brake shoe begins to move. When the brake shoe is fully braked, the brake arm presses the microswitch contacts, and the switch changes from the open state to the closed state. The acquisition module immediately records the system timestamp at this time, corresponding to T1 (left brake shoe) and T2 (right brake shoe), respectively. When braking ends and the brake shoe resets, the switch returns to the open state, and the acquisition module synchronously records the reset time to determine whether the switch reset is normal. During the acquisition process, the time difference between T1, T2 and the braking command time and the inflection point time of the speed curve is recorded to generate timing action data. After binding the timestamp, the data is archived synchronously with other signals to ensure that fault verification can be achieved through timing comparison later.

[0043] This embodiment uses a KEYENCEIL-600 series non-contact laser displacement sensor with a measurement range of 0-5mm, a measurement accuracy of ≤±0.001mm, a sampling frequency of 100Hz, and synchronization with the system timestamp. The sensor is installed on one side of the brake disc, with the probe facing the gap between the brake disc and the brake shoe. The installation height is flush with the brake shoe to ensure that the movement of the brake shoe during braking does not obstruct the probe. The installation location should be far away from dusty and oily areas, and the probe surface should be cleaned regularly to avoid affecting the measurement accuracy. At the same time, vibration-proof fixation should be ensured to reduce the interference of traction sheave vibration on the sensor.

[0044] The brake clearance displacement data in this embodiment represents the size and real-time fluctuation of the gap between the brake disc and the brake shoes. It is a core parameter for identifying potential problems such as brake shoe wear, excessive brake clearance, and loose brake shoes. The normal braking clearance range of an elevator brake is 0.5-1.0mm. During braking, the brake shoes gradually come into contact with the brake disc, and the gap should decrease uniformly to 0 (the brake shoes and brake disc are completely in contact). The gap remains stable when the machine is stopped, with a fluctuation range of ≤±0.01mm. If the gap exceeds the normal range, the fluctuation is too large, or the gap decreases unevenly during braking, it indicates potential problems such as excessive brake shoe wear, improper brake clearance adjustment, and loose brake shoes. These problems can lead to a decrease in braking performance and even safety risks.

[0045] The laser displacement sensor emits laser signals into the gap between the brake disc and the brake shoe in real time. It detects the gap distance through the principle of laser reflection and records a displacement acquisition point every 10ms, generating time-series data of "timestamp + displacement value" and binding it to the system's unified timestamp. During the acquisition process, a 5-point sliding average filter is used to filter out instantaneous abnormal displacement values ​​caused by dust and vibration, and retain the effective displacement signal. At the same time, the speed signal is synchronously correlated to record the gap displacement value at key moments such as the start of braking, the braking position, and the steady state after stopping.

[0046] S120 achieves the fusion of the first wheel speed signal and the second wheel speed signal based on a preset fusion model. Based on this, such as Figure 2 As shown, S120 specifically includes the following steps:

[0047] S121: The fusion model filters out the noise component in the first wheel speed signal to obtain the processed wheel speed signal;

[0048] S122: Based on a preset algorithm, multiple anchor points are extracted from all the second acquisition points. The anchor point is one of the second acquisition points in the stable interval of the second wheel speed signal. When the elevator runs in the time period corresponding to the steady-state interval, the elevator is in a uniform speed state.

[0049] S123: Based on all the anchor points and the first acquisition point corresponding to the timestamp, determine the vibration drift of each anchor point to the first acquisition point corresponding to the timestamp;

[0050] S124: Based on the vibration drift, correct the first wheel speed signal to obtain a first corrected wheel speed signal;

[0051] S125: Based on the first corrected wheel speed signal, obtain the third acquisition point where the timestamp of the second wheel speed signal is missing relative to the first corrected wheel speed signal, so as to obtain the second corrected wheel speed signal;

[0052] S126: By using a preset fusion strategy, the first corrected wheel speed signal and the second corrected wheel speed signal are fused to obtain the fused wheel speed signal.

[0053] In this embodiment, S121 achieves noise component filtering in the following manner:

[0054] The first wheel speed signal is decomposed by wavelet to obtain a low-frequency component and several thousand high-frequency components; abnormal noise in the high-frequency component is removed to obtain the effective high-frequency component; the low-frequency component and the effective high-frequency component are reconstructed in reverse to obtain the processed wheel speed signal.

[0055] This embodiment performs wavelet denoising on the first wheel speed signal (containing time-series data from several first acquisition points, mixed with noise such as radial vibration and low-frequency vibration of the traction wheel) acquired by the contact sensor to obtain the processed wheel speed signal. For example, the wavelet denoising method can be as follows: First, a db4 wavelet basis (adapting to the non-stationary characteristics of traction wheel speed signals, possessing both good temporal locality and frequency resolution, and strong anti-vibration interference capability) is selected to perform a 5-level wavelet decomposition on the first wheel speed signal. After decomposition, one low-frequency component and five high-frequency components are obtained. The low-frequency component corresponds to the actual speed change trend of the traction wheel and is unaffected by vibration interference. The five high-frequency components are all clutter signals generated by traction wheel vibration, electromagnetic interference, sensor errors, etc. Then, the 3σ criterion is used to remove abnormal clutter from all high-frequency components. First, the mean and standard deviation of each high-frequency component signal are calculated, and the clutter judgment threshold is determined as [mean - 3σ, mean + 3σ, ...]. [3σ] High-frequency signal points exceeding the threshold are identified as sudden abnormal noise such as instantaneous vibration impact and electromagnetic pulse and are completely eliminated. The effective high-frequency components within the threshold range (reasonable signals from slight vibrations, avoiding excessive stripping that would distort the speed signal) are retained. Finally, inverse wavelet transform is used to strictly match the parameters during wavelet decomposition (db4 wavelet basis, 5-level decomposition). The processed low-frequency components and effective high-frequency components are reconstructed in reverse. During the reconstruction process, the true trend of the traction wheel speed is fully restored. Finally, a smooth and continuous wheel speed signal without obvious vibration spikes or abrupt jumps is obtained, providing high-quality basic data for subsequent anchor point extraction, vibration drift calculation, and correction of the first wheel speed signal.

[0056] Understandably, based on the frequency division principle of wavelet decomposition, a sampling frequency of 1000Hz corresponds to a Nyquist frequency of 500Hz. Each decomposition layer divides the frequency range of the current signal into two equal parts: a low-frequency component (approximation coefficient) and a high-frequency component (detail coefficient) corresponding to 0-250Hz. The second decomposition layer further divides the low-frequency component from the first layer, resulting in a low-frequency component of 0-125Hz and a high-frequency component of 125-250Hz. The third layer yields a low-frequency component of 0-62.5Hz and a high-frequency component of 62.5-125Hz. The fourth layer yields a low-frequency component of 0-31.25Hz and a high-frequency component of 31.25-62.5Hz. The fifth layer finally yields a low-frequency component of 0-31.25Hz and five high-frequency components corresponding to 31.25-62.5Hz, 62.5-125Hz, 125-250Hz, and 250-500Hz, respectively. The low-frequency components of 0-31.25Hz completely cover the speed change frequency of the elevator traction sheave during normal operation and braking (usually ≤10Hz), corresponding to the actual speed trend of the traction sheave; all high-frequency components above 31.25Hz are useless noise such as radial vibration of the traction sheave, electromagnetic interference of the control cabinet, and sensor noise, which matches the physical meaning of "low frequency is effective signal and high frequency is interference noise" in this embodiment.

[0057] This embodiment can separate the low-frequency effective component representing the actual operating law of the traction wheel in the first wheel speed signal of the contact sensor from the high-frequency noise component generated by radial vibration, low-frequency disturbance, and electromagnetic interference. By eliminating sudden abnormal noise in the high-frequency component and retaining the effective high-frequency component with reasonable small fluctuations, it not only filters out the glitches, jumps, and pulse interference caused by mechanical vibration of the sensor, but also avoids the excessive smoothing of speed characteristics and signal distortion caused by global filtering. It retains the wheel speed timing change trend and working condition details. The reconstructed wheel speed signal has high smoothness, high fidelity, and high stability, effectively suppressing the data distortion problem caused by long-term vibration of the contact sensor.

[0058] S122 of this embodiment specifically includes the following steps:

[0059] Calculate the volatility and average velocity of all second acquisition points in each sliding window of a preset length, wherein the sliding window slides from the starting point to the ending point on the time axis on the second wheel speed signal; determine all stable intervals based on the volatility and average velocity, wherein the volatility of all sliding windows corresponding to the stable interval is less than a second volatility threshold, and the difference in average velocity between two adjacent sliding windows in all sliding windows corresponding to the stable interval is less than a first difference threshold; extract the second acquisition point located in the middle of each stable interval as the anchor point.

[0060] The definitions of S122 in this embodiment are as follows:

[0061] 1. Sliding window: refers to setting a fixed preset time length (e.g., 50ms) on the time axis of the second wheel speed signal to collect the second collection points within this time range in batches. It can slide continuously along the time axis from the signal start point to the end point to realize the segmented analysis of the second wheel speed signal. In this embodiment, the step size of the sliding window is 1.

[0062] 2. The second wheel speed signal is time-series speed data acquired by a non-contact sensor, which includes several discrete second acquisition points.

[0063] 3. The second acquisition point is each discrete speed acquisition data point in the second wheel speed signal, with a unique timestamp, corresponding to the non-contact speed measurement value at a certain moment.

[0064] 4. Volatility is the difference between the maximum and minimum values ​​among all second-level sampling points within a single sliding window, divided by the average velocity of that window. It is used to quantify the degree of velocity fluctuation within the sliding window and reflect velocity stability.

[0065] 5. Average speed represents the arithmetic mean of the speed values ​​of all second-collection points within a single sliding window, reflecting the overall speed level during that time period.

[0066] 6. The stable interval represents the continuous time period in the second round speed signal that satisfies "all sliding window volatility is less than the second volatility threshold" and "the average speed difference between adjacent sliding windows is less than the first difference threshold", corresponding to the elevator's uniform speed operation state.

[0067] 7. The second fluctuation threshold represents the preset speed fluctuation judgment standard (e.g., ±2%), which is used to distinguish whether the speed within the window is stable. If it is less than the threshold, it means that the speed fluctuation within the window is small and the state is stable.

[0068] 8. The first difference threshold represents the preset standard for judging the average speed difference between adjacent windows (e.g., ±0.01m / s), which is used to distinguish whether the speed between adjacent time periods is continuous and stable. If it is less than this threshold, it means that there is no obvious jump in speed.

[0069] 9. Anchor point refers to a second acquisition point selected from the middle of each stable interval, which serves as a high-precision velocity reference point for subsequent vibration drift calculation and correction of the first wheel speed signal.

[0070] In this embodiment, a fixed-length sliding window is first set for the second round speed signal. This window slides uniformly along the time axis from the signal's starting point to its ending point. Each time the window slides, the volatility and average speed of all second acquisition points within the current sliding window are calculated, and the segmented calculation of the entire second round speed signal is completed sequentially. Subsequently, based on the preset second volatility threshold and first difference threshold, all continuous time periods that meet the conditions are selected as stable intervals—that is, the volatility of each sliding window within this time period is less than the second volatility threshold (indicating small speed fluctuations within the window), and the average speed difference between two adjacent sliding windows is less than the first difference threshold (indicating that the speeds of adjacent time periods are continuous and without jumps). Finally, for each selected stable interval, the second acquisition point at the middle position of the time axis of that interval is selected and determined as the anchor point, completing the extraction of all anchor points.

[0071] In this embodiment, the overall accuracy of the second wheel speed signal collected by the non-contact sensor is insufficient. However, when the elevator is running at a constant speed, the traction sheave vibration is minimal and the speed is most stable, resulting in minimal fluctuation of the corresponding second wheel speed signal. Therefore, by using a sliding window segmented analysis and employing two thresholds—the volatility and the average speed difference between adjacent windows—the stable interval corresponding to the elevator's constant speed operation can be selected, eliminating unstable signals from non-uniform speed segments (acceleration, deceleration, and braking). Selecting the second acquisition point in the middle of the stable interval as the anchor point avoids potential edge errors at the start and end points of the interval (such as speed fluctuations during interval transitions), ensuring that the speed value of the anchor point is closest to the true uniform speed of the stable interval, and maximizing the high-precision advantage of the non-contact sensor in the uniform speed segment.

[0072] Based on this, the preset algorithm used in this embodiment traverses the non-contact second-round speed time-series signal segment by segment through a fixed-duration sliding window. It sequentially calculates the speed fluctuation rate within the window and performs interval average speed statistics. Combined with dual threshold conditions, it identifies stable operating conditions and filters stable intervals, selecting feature collection points in the middle of each stable interval as anchor points. This algorithm objectively identifies the signal intervals corresponding to the elevator's uniform speed operation by quantifying the magnitude of speed fluctuations and the speed continuity between adjacent time periods, thus achieving automatic and standardized extraction of steady-state anchor points.

[0073] S123 specifically includes the following steps:

[0074] Using the system's unified timestamp as the matching benchmark, each selected non-contact anchor point is used as a reference benchmark. Following the principle of one-to-one correspondence between timestamps, the first acquisition point that is completely consistent with the anchor point's time is matched among all the first acquisition points of the first wheel speed signal after wavelet denoising. Using the anchor point's speed value as the standard, the measured speed value of the corresponding first acquisition point at the same timestamp is read. By calculating the speed difference between the two, the measurement offset caused by the radial vibration and low-frequency vibration interference of the traction wheel to the contact sensor at that moment is quantified, which is the single-point vibration drift. After traversing all anchor points to complete the point-by-point timestamp matching and difference calculation, multiple sets of vibration drift data with discrete distribution and time sequence correspondence are formed, providing a quantitative correction basis for subsequent segmented correction of the first wheel speed signal.

[0075] Assuming the timestamp of the first steady-state anchor point is 1000ms, the corresponding non-contact velocity value is 2.00m / s. In the processed first wheel speed signal, the measured velocity of the first acquisition point with the same timestamp of 1000ms is found to be 2.04m / s. Subtracting the standard velocity of the anchor point from the velocity of the first acquisition point yields a vibration drift of 0.04m / s at that moment. Then, several other anchor points with timestamps of 1500ms, 2000ms, etc., are selected, and the velocities of the first acquisition points at the corresponding moments are matched. The vibration drift at each time point is calculated sequentially, ultimately forming a set of vibration drift samples arranged by time, which are used for subsequent segmented compensation and correction of the entire first wheel speed signal.

[0076] Based on this, S124 is specifically as follows:

[0077] First, all anchor points that have completed timestamp matching and vibration drift calculation are arranged in ascending order according to their temporal sequence. Using the time position of each anchor point as the boundary node, the complete first-round velocity time-series signal after wavelet denoising is divided into several intermediate correction intervals enclosed by two adjacent anchor points. Simultaneously, two edge correction intervals are defined: one before the first anchor point and one after the last anchor point. For each intermediate interval formed by adjacent anchor points, combining the time difference between the start and end anchor points and the corresponding vibration drift difference, a linear interpolation algorithm is used to characterize and fit the continuous vibration drift value at each first acquisition point within the interval, following the linear time variation law. This realistically restores the actual characteristic of the contact sensor's vibration drift gradually changing with the running sequence. For the two edge intervals before and after, since interpolation fitting cannot be performed due to the lack of double-sided anchor points, the vibration drift value of the single anchor point closest to the edge interval is taken as the uniform fixed compensation drift amount for the entire edge interval. After completing the fitting and completion of the vibration drift values ​​at all moments in the full time series, the correction calculation is performed on each first acquisition point in the first wheel speed signal. The original measured speed value of each acquisition point is subtracted from the fitted vibration drift value corresponding to the current moment to offset the measurement offset error caused by the radial vibration and low-frequency vibration of the traction wheel to the contact sensor. After all acquisition points are corrected, all corrected speed data are re-integrated and spliced ​​according to the time series to finally obtain the first corrected wheel speed signal with overall elimination of vibration interference, continuous and stable time series, and significantly improved measurement accuracy.

[0078] Assuming that the initial matching calculation yields three time-series anchor points with timestamps of 1000ms, 2000ms, and 3000ms respectively, and corresponding vibration drift values ​​of 0.04m / s, 0.06m / s, and 0.05m / s respectively; firstly, using these three anchor points as boundaries, the first wheel speed signal is divided into the front edge interval of 0-1000ms, the middle interval of 1000-2000ms, the middle interval of 2000-3000ms, and the rear edge interval after 3000ms. Within the 1000ms to 2000ms interval, spanning 1000ms, the vibration drift gradually and uniformly changes from 0.04m / s to 0.06m / s. Taking the midpoint of the interval at 1500ms, the vibration drift at that moment can be accurately calculated as 0.05m / s through linear interpolation. If the original acquisition velocity of the first wheel at that moment is 2.05m / s, subtracting the corresponding drift value of 0.05m / s from the original velocity yields the corrected standard velocity of 2.00m / s. Similarly, at 2500ms within the 2000ms to 3000ms interval, the vibration drift is calculated to be 0.055m / s through interpolation. If the original velocity is 2.12m / s, the corrected velocity is 2.065m / s. For all acquisition points within the range of 0–1000 ms, a fixed drift value of 0.04 m / s at the nearest 1000 ms anchor point is used for correction. For all acquisition points after 3000 ms, a point-by-point compensation correction is performed using 0.05 m / s at the 3000 ms anchor point. Following the interpolation and edge-fixed compensation rules described above, drift subtraction correction is performed on all time-series acquisition points in the first wheel speed signal. Finally, all corrected data are stitched together in time sequence to generate a complete, smooth, and vibration-displacement-free first corrected wheel speed signal.

[0079] Based on this, S125 specifically includes the following steps:

[0080] Using the first corrected wheel speed signal, which has undergone vibration drift correction, has a complete timing sequence, and a uniform sampling interval, as a unified time reference template, its entire standard timestamp sequence is first extracted as a reference scale. Then, it is compared and verified one by one with the timestamps of all the second acquisition points inherent in the original second wheel speed signal. This process identifies gaps in the standard timing sequence where acquisition points are not deployed in the second wheel speed signal. These gaps are then marked as locations where third acquisition points need to be added. For each gap, the system automatically retrieves the two closest valid second acquisition points before and after that moment, extracts the timestamp interval between the two points and the corresponding measured speed value, and uses time as a reference. The linear correlation law employs an interpolation algorithm to accurately fit the equivalent speed value corresponding to each missing timestamp according to the time proportion, thereby generating a third acquisition point with a standard timestamp and a fitted speed value. Finally, all the original second acquisition points of the second wheel speed signal and all the third acquisition points generated by interpolation are summarized and integrated, and then reordered and organized according to the unified timestamp of the system from smallest to largest. Any duplicate timestamp data is eliminated, and all missing time sequence nodes are filled in. Finally, a second corrected wheel speed signal with the same sampling interval as the first corrected wheel speed signal, completely aligned timestamps, continuous throughout without breaks, and uniform time sequence distribution is obtained, laying the foundation for time sequence matching for subsequent dual wheel speed signal fusion.

[0081] Assuming the first corrected wheel speed signal is sampled at fixed 10ms intervals, it possesses a complete and continuous standard timestamp sequence of 10ms, 20ms, 30ms, 40ms, and 50ms, with a regular timing sequence and no gaps or breaks; however, the original second wheel speed signal acquired by the non-contact sensor is limited by its own sampling mechanism, with only valid second acquisition points at 10ms, 30ms, and 50ms, corresponding to a speed value of 1.75m / s. There is no acquired data at the standard time points of 20ms and 40ms, indicating significant timestamp gaps and sampling discontinuities. Using the complete timestamps of the first corrected wheel speed signal as a reference, the 20ms and 40ms timestamps are accurately identified. To fill the missing positions of the third sampling points, two valid second sampling points adjacent to each missing position were selected. According to the time-series linear interpolation rules, combined with the time span and velocity values ​​of the preceding and following points, the velocities corresponding to the 20ms and 40ms times were calculated and fitted to be 1.75m / s, generating two compliant third sampling points. Then, the original 10ms, 30ms, and 50ms second sampling points of the second wheel speed signal were rearranged and merged with the newly interpolated 20ms and 40ms third sampling points according to the time sequence to fill in all missing time nodes. Finally, a second corrected wheel speed signal with uniform sampling interval, full timestamp coverage, and complete synchronization with the time sequence of the first corrected wheel speed signal was generated.

[0082] Based on this, S126 specifically includes the following steps:

[0083] Align the first and second corrected wheel speed signals on the timestamp; determine a weight combination based on the stable interval, wherein when the first acquisition point and the corresponding second or third acquisition point are located within the stable interval, the weight combination includes a first weight coefficient for the first acquisition point and a second weight coefficient for the corresponding second or third acquisition point; when the first acquisition point and the corresponding second or third acquisition point are not located within the stable interval, the weight combination includes a third weight coefficient for the first acquisition point and a fourth weight coefficient for the corresponding second or third acquisition point, wherein the third weight coefficient is greater than the first weight coefficient and the fourth weight coefficient is less than the second weight coefficient; obtain a fused wheel speed signal based on the third weight coefficient, the first weight coefficient, the fourth weight coefficient, the second weight coefficient, the first corrected wheel speed signal, and the second corrected wheel speed signal.

[0084] In this embodiment, the first and second corrected wheel speed signals are first precisely aligned in terms of timestamps. Since the second corrected wheel speed signal has filled in all the missing third acquisition points in the timestamps and is completely consistent with the sampling interval and timestamp sequence of the first corrected wheel speed signal, it is only necessary to match point by point according to the unified timestamp to ensure that each moment corresponds to a set of "first corrected wheel speed acquisition point (after first acquisition point correction) + second corrected wheel speed acquisition point (second acquisition point or third acquisition point)". Then, all stable intervals determined in the previous anchor point extraction stage are retrieved to determine whether the acquisition point corresponding to each timestamp is in a stable interval, and then the corresponding weight combination is determined: if the first acquisition point and the corresponding second acquisition point (or third acquisition point) at that moment are in a stable interval, the first weight combination is adopted, that is, the first acquisition point is assigned the first weight combination. A weighting coefficient (e.g., 0.3) is assigned to the first acquisition point (or the third acquisition point), and a second weighting coefficient (e.g., 0.7) is assigned to the second acquisition point (or the third acquisition point). If the acquisition point at that moment is not within the stable interval, a second weighting combination is used, that is, the first acquisition point is assigned a third weighting coefficient (e.g., 0.7), and the corresponding second acquisition point (or the third acquisition point) is assigned a fourth weighting coefficient (e.g., 0.3), where the third weighting coefficient is greater than the first weighting coefficient and the fourth weighting coefficient is less than the second weighting coefficient. Finally, for the two sets of corrected velocity values ​​corresponding to each timestamp, a weighted sum is calculated according to the corresponding weighting combination (fusion velocity = first corrected wheel speed value × corresponding weighting coefficient + second corrected wheel speed value × corresponding weighting coefficient). After completing the fusion calculation point by point, all fused velocity values ​​are integrated in time sequence to obtain the complete fused wheel speed signal.

[0085] The mechanism of this fusion scheme is based on the performance advantages of dual sensors under different operating conditions. It achieves complementary advantages through dynamic weight allocation and avoids the inherent defects of a single sensor. The stable range corresponds to the elevator's uniform speed operation, where the traction sheave vibration is minimal. The second acquisition point (or the supplemented third acquisition point) collected by the non-contact sensor has the highest accuracy. While the contact sensor, although corrected, still has slight residual vibration errors, a higher weight is assigned to the non-contact acquisition point (second weight coefficient), and a lower weight is assigned to the contact acquisition point (first weight coefficient), maximizing the high accuracy advantage of the non-contact sensor in steady-state conditions. The unstable range (acceleration, deceleration, and braking stages) corresponds to the elevator's drastic speed changes and severe traction sheave vibration. Due to limitations in the measurement principle, the accuracy of the non-contact sensor drops significantly. However, the contact sensor, after vibration drift correction, possesses good temporal continuity and anti-interference capabilities, accurately capturing dynamic speed changes. Therefore, a higher weight is assigned to the contact acquisition point (third weight coefficient), and a lower weight is assigned to the non-contact acquisition point (fourth weight coefficient), fully leveraging the continuity advantage of the contact sensor in dynamic conditions. This dynamic weight allocation achieves optimal fusion of the two corrected signals, balancing the high accuracy and temporal continuity of the fused wheel speed signal.

[0086] In this embodiment, after obtaining the fused wheel speed signal, it is necessary to correct the end portion of the fused wheel speed signal. The specific correction method is as follows:

[0087] Based on a preset tool, the brake engagement time and stop time of the fused wheel speed signal are extracted; the attenuation slope and speed fluctuation of the fused wheel speed signal at the corresponding parts of the brake engagement time and stop time are calculated; when the attenuation slope is greater than a preset slope, or the speed fluctuation is greater than a first fluctuation threshold, the corresponding part is corrected so that the attenuation slope of the corresponding part is less than or equal to the preset slope, and the speed fluctuation is less than or equal to the first fluctuation threshold.

[0088] This embodiment utilizes a preset signal analysis tool (preset tool) to analyze the acquired fused wheel speed signal segment by segment, accurately extracting the key time nodes at the end of the fused wheel speed signal—the moment the brake is engaged (i.e., the moment the brake is fully closed and begins to exert its braking effect) and the moment the elevator stops (i.e., the moment the elevator completely stops running). Subsequently, the signal analysis tool calculates the attenuation slope of the fused wheel speed signal between these two moments (i.e., the rate at which the speed decreases over time), and simultaneously calculates the speed fluctuation amplitude (i.e., the difference between the maximum and minimum speed values) within this time period. Next, the calculated attenuation slope is compared with a preset slope threshold, and the speed fluctuation amplitude is compared with a preset first fluctuation threshold. If the attenuation slope is greater than the preset slope threshold, or the speed fluctuation amplitude is greater than the first fluctuation threshold, then the segment of the fused wheel speed signal is corrected by linearly adjusting to reduce the attenuation slope and smooth the speed fluctuation until the attenuation slope is ≤ the preset slope threshold and the speed fluctuation amplitude is ≤ the first fluctuation threshold. After the correction is completed, the attenuation slope and fluctuation amplitude of the signal segment are verified again to ensure that they meet the preset standards, thus completing the correction of the end of the fused wheel speed signal.

[0089] After the brake engages, the elevator should smoothly decelerate from its constant speed to a stop. The speed decay should be gradual and without drastic fluctuations. If the decay slope is too large (speed decreases too quickly) or the fluctuation amplitude is too large, it indicates that vibration interference from the contact sensors still exists, or that there is a sampling deviation in the non-contact sensors, which can lead to misdiagnosis in subsequent fault diagnosis. Therefore, by extracting the brake engagement time and the stopping time, calculating the speed decay slope and fluctuation amplitude in this interval, and comparing them with a preset threshold, linear correction can be applied to signals exceeding the threshold. This effectively offsets the speed deviation caused by vibration interference, ensuring that the fused wheel speed signal is consistent with the actual operating state of the elevator, and providing accurate speed data support for subsequent fault diagnosis.

[0090] In S130, the fused wheel speed signal needs to be fed into the deep learning model to obtain the confidence of all first fault types pre-stored in the deep learning model.

[0091] In this embodiment, the deep learning model employs a feature fusion network that combines a one-dimensional convolutional neural network (1D-CNN) and a bidirectional long short-term memory network (BiLSTM) in parallel with an attention mechanism. The 1D-CNN is used to extract local morphological features from the fused wheel speed signals, such as speed fluctuations and transient spikes during braking. The BiLSTM is used to capture the temporal dependencies throughout the braking process, such as the interaction between the forward power-building phase and the backward brake-release phase. The attention mechanism automatically focuses on the most discriminative critical moments during braking, thereby achieving high-precision classification of multiple fault types.

[0092] The deep learning model is trained offline using a supervised approach. The training dataset consists of historically measured fused wheel speed signal samples and corresponding real fault labels. The samples cover different fault types and normal operating states, and data augmentation techniques are used to simulate signal drift of the traction machine brake under harsh conditions such as low air pressure, low temperature, and high humidity. Cross-entropy is used as the loss function during training, and the Adam optimizer is employed. Label smoothing regularization and early stopping mechanisms are introduced to prevent overfitting. After training, the model is embedded as an inference module. During online inference, only forward computation is performed, outputting the confidence score for each first fault type.

[0093] The input to the deep learning model is a fixed-length fused wheel speed signal time series. First, it passes through a one-dimensional convolutional module containing three cascaded convolutional blocks. Each block consists of a one-dimensional convolutional layer with 32 kernels, a batch normalization layer, and a ReLU activation function, and is progressively reduced in dimensionality through max pooling. In parallel, the input simultaneously enters a BiLSTM module containing two stacked layers, each with 64 hidden units, outputting a complete temporal state sequence. The two feature streams are concatenated and fed into a multi-head self-attention layer, dynamically weighted according to the criticality of the braking process. Subsequently, they are compressed into a fixed-dimensional feature vector through a global average pooling layer. Finally, a fully connected layer and a Softmax layer output a probability distribution whose number equals the total number of the first fault types, with each probability value representing the confidence level of the corresponding fault type.

[0094] The knowledge graph in S140 can be constructed based on the following method, specifically:

[0095] Using historical maintenance data, fault case database, and bench test data of elevator brakes as data sources, fault type entity nodes are first defined. Each entity node includes attributes such as fault name, cause, and handling method. Standard electrical parameter signal curves, standard microswitch action timing, and standard displacement signal curves collected under the corresponding fault conditions are stored as built-in feature templates. Simultaneously, a "signal feature-fault" relationship is established, using expert experience to semantically associate fault modes such as brake jamming, spring fatigue, and residual magnetism decay with signal features such as peak distortion of starting current, action timing deviation, and abnormal displacement rebound. The structured data is stored in a graph database, forming a knowledge graph with fault type nodes at its core and external standard multi-source timing signal templates. When the deep learning model outputs the fault type with the highest confidence, the first standard electrical parameter signal, first standard microswitch signal, and first standard displacement signal linked to that fault type node can be directly extracted through graph retrieval for subsequent consistency comparison.

[0096] In S150, the three deviations are calculated as follows:

[0097] Electrical parameter signal deviation: First, check the actual electrical parameter signal. With the first standard electrical parameter signal Z-score normalization is performed, and then the normalized distance between the two is calculated using the dynamic time warping algorithm.

[0098]

[0099] The larger the distance, the greater the difference in waveform shape. To map the deviation to a uniform scale within the [0,1] interval, the following steps are taken: Divide by the self-regulating distance of the standard signal The ratio of the electrical parameter deviation to the signal length N yields the normalized electrical parameter deviation:

[0100]

[0101] This is a very small constant to prevent the denominator from being 0.

[0102] Microswitch signal deviation: The microswitch signal is abstracted as an ordered sequence of events containing the action type (e.g., on / off) and the timing of the action. Let the actual event sequence be... The first standard event sequence is We use dynamic programming to align the events of the two sequences and define the matching cost function after alignment:

[0103]

[0104] Where M is the maximum number of events covered by the alignment. This is an indicator function (counted as 1 if the action type is different, otherwise counted as 0). This is the time offset penalty coefficient. The cost is then converted into bias through exponential mapping:

[0105]

[0106] The scaling factor is such that the deviation is 0 when the cost is 0, and the deviation approaches 1 when the cost increases.

[0107] Displacement signal deviation: relative to the actual displacement signal With the first standard displacement signal The same method is used. After normalization, the dynamic time-warped distance is calculated and normalized in the same way:

[0108]

[0109] The first consistency deviation is obtained by weighting and fusing the above three deviations according to preset weights.

[0110] The specific steps in S160 are as follows:

[0111] Based on the knowledge graph, a second fault type is determined corresponding to the second standard electrical parameter signal, the second standard microswitch signal, and the second standard displacement signal. The second consistency deviation, determined based on the deviations between the second standard electrical parameter signal and the actual electrical parameter signal, the second standard microswitch signal and the actual microswitch signal, and the second standard displacement signal and the actual displacement signal, is less than a deviation threshold. The confidence level of a first fault type that is the same as the second fault type is obtained. When the confidence level of the first fault type that is the same as the second fault type is greater than or equal to the confidence level threshold, the third initial fault type is output as an abnormal result. When the confidence level of the first initial fault type that is the same as the third initial fault type is less than the confidence level threshold, both the second fault type and the first initial fault type are output as abnormal results.

[0112] It is understandable that the deviation in S160 is calculated in the same way as the deviation in S150.

[0113] In this embodiment, when the first consistency deviation of the fault type with the highest confidence output of the deep learning model based on the fused wheel speed signal exceeds the threshold, it indicates that there is a significant deviation between the actual multi-source physical signal and the standard fault mode "considered" by the deep learning. This deviation may originate from local perturbations in the wheel speed signal, fuzzy boundaries of specific faults in the deep model, or the actual occurrence of a compound fault. Based on this, the system determines that the diagnostic results of a single deep learning path cannot be fully trusted, thereby triggering a secondary verification process based on the knowledge graph, using the actual form of the physical signal as an anchor point to re-search for the most matching fault type.

[0114] In this embodiment, the pre-defined "fault type-standard signal template" link relationship in the knowledge graph is utilized to traverse all known faults. For each candidate fault, the normalized deviation between the actual electrical parameters, microswitch, displacement signal, and their standard template is calculated, and the deviations are fused to obtain the second consistency deviation of the candidate fault. Among all candidate faults, the fault with the smallest deviation value that is lower than the preset deviation threshold is selected as the second fault type. This represents a diagnostic conclusion that highly matches a certain type of fault mode in the knowledge base, based solely on the measured waveforms of the brake's electrical and mechanical actions and physical displacement, without relying on wheel speed signals.

[0115] After obtaining the second fault type, the system backtracks to the complete confidence vector previously generated by the deep learning model and extracts the confidence value of the fault type that matches the second fault type. If this confidence value is greater than or equal to a preset confidence threshold, it indicates that although there was a deviation between the previous wheel speed signal and the standard template, the deep learning model actually gave a high degree of confidence in this fault that highly matches the physical waveform. The two independent paths resonate in terms of fault category, verifying the authenticity of the fault. Therefore, the system outputs it as the only abnormal result, and the reliability of the diagnosis is cross-validated by the two paths.

[0116] If the confidence level of the first fault type corresponding to the second fault type is lower than the threshold, it means that the deep learning model does not regard this fault, which highly matches the physical signal, as a high-probability event. This may indicate that the model has insufficient feature learning for this type of fault, or that the wheel speed signal itself fails to adequately characterize the fault. In this case, the system will not favor a single path but will output both diagnostic results simultaneously: outputting the second fault type to reflect the true direction of the actual physical signal, and simultaneously outputting the first initial fault type to retain the deep learning model's judgment based on the wheel speed signal. The joint anomaly results aim to present maintenance personnel with a complete picture of the two possible fault states of the brake, guiding further on-site investigation and comprehensive assessment.

[0117] This application provides an electronic device that may include a processing module and a memory. The memory stores a computer program, which, when executed by the processor, enables the electronic device to perform the corresponding steps in the aforementioned brake malfunction diagnosis method.

[0118] In this embodiment, the processor can be an integrated circuit chip with signal processing capabilities. For example, the processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0119] The memory can be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, etc.

[0120] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic device described above can be referred to the corresponding steps in the aforementioned method, and will not be elaborated further here.

[0121] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the brake anomaly diagnosis method as described in the above embodiments.

[0122] Computer-readable storage media may be magnetic disks, optical disks, read-only memory, random access memory, flash memory, USB flash drives, hard disks, or solid-state drives, etc., and may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implement the methods shown in the above embodiments.

[0123] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned brake anomaly diagnosis method. The computer program product may exist in a computer-readable storage medium in forms including, but not limited to, source files, executable files, and installation package files.

[0124] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, electronic device, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0125] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0126] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for diagnosing brake malfunctions, characterized in that, The method includes: The system acquires a first wheel speed signal, a second wheel speed signal, an actual electrical parameter signal, an actual micro-switch signal, and an actual displacement signal. The first wheel speed signal and the second wheel speed signal are obtained based on a contact sensor and a non-contact sensor, respectively. Both the first wheel speed signal and the second wheel speed signal are time-series data. The first wheel speed signal includes several first acquisition points, and the second wheel speed signal includes several second acquisition points. The first wheel speed signal and the second wheel speed signal are input into the fusion model, and the fusion model fuses the first wheel speed signal and the second wheel speed signal to obtain the fused wheel speed signal; The fused wheel speed signal is input into a deep learning model, and the confidence level of each first fault type is output through the deep learning model. The number of first fault types is greater than or equal to 2, and all first fault types are stored in the deep learning model. Based on the fault type with the highest confidence level, and using a knowledge graph, determine the first standard electrical parameter signal, the first standard micro switch signal, and the first standard displacement signal corresponding to the fault type with the highest confidence level in the knowledge graph. Based on the deviations between the actual electrical parameter signal and the first standard electrical parameter signal, the deviations between the actual microswitch signal and the first standard microswitch signal, and the deviations between the actual displacement signal and the first standard displacement signal, a first consistency deviation is determined. When the first consistency deviation is greater than the deviation threshold, an abnormal result is determined based on the knowledge graph and the preset strategies for all first fault types.

2. The method according to claim 1, characterized in that, The predetermined strategy based on the knowledge graph and all first fault types determines the abnormal results, including: Based on the knowledge graph, the second fault type corresponding to the second standard electrical parameter signal, the second standard micro switch signal, and the second standard displacement signal is determined, wherein the second consistency deviation determined based on the deviation between the second standard electrical parameter signal and the actual electrical parameter signal, the deviation between the second standard micro switch signal and the actual micro switch signal, and the deviation between the second standard displacement signal and the actual displacement signal is less than the deviation threshold. Obtain the confidence level of the first fault type that is the same as the second fault type; When the confidence level of the first fault type is greater than or equal to the confidence threshold, the third initial fault type is output as an abnormal result. When the confidence level of the first initial fault type is less than the confidence threshold, both the second fault type and the first initial fault type are output as abnormal results.

3. The method according to claim 1, characterized in that, The step of inputting the first wheel speed signal and the second wheel speed signal into the fusion model, wherein the fusion model fuses the first wheel speed signal and the second wheel speed signal to obtain a fused wheel speed signal, includes: The fusion model filters out the noise components in the first wheel speed signal to obtain the processed wheel speed signal; Based on a preset algorithm, multiple anchor points are extracted from all the second collection points. The anchor point is one of the second collection points in the stable interval of the second wheel speed signal. When the elevator runs in the time period corresponding to the steady-state interval, the elevator is in a uniform speed state. Based on all the anchor points and the corresponding first acquisition point on the timestamp, determine the vibration drift of each anchor point relative to the corresponding first acquisition point on the timestamp; Based on the vibration drift, the first wheel speed signal is corrected to obtain a first corrected wheel speed signal; Based on the first corrected wheel speed signal, a third acquisition point where the timestamp of the second wheel speed signal is missing relative to the first corrected wheel speed signal is obtained, so as to obtain the second corrected wheel speed signal; By using a preset fusion strategy, the first corrected wheel speed signal and the second corrected wheel speed signal are fused to obtain the fused wheel speed signal.

4. The method according to claim 3, characterized in that, The step of fusing the first corrected wheel speed signal and the second corrected wheel speed signal through a preset fusion strategy to obtain the fused wheel speed signal includes: Align the first and second corrected wheel speed signals on the timestamp; Based on the stable interval, a weight combination is determined. When the first sampling point and the corresponding second or third sampling point are located within the stable interval, the weight combination includes a first weight coefficient for the first sampling point and a second weight coefficient for the corresponding second or third sampling point. When the first sampling point and the corresponding second or third sampling point are not located within the stable interval, the weight combination includes a third weight coefficient for the first sampling point and a fourth weight coefficient for the corresponding second or third sampling point. The third weight coefficient is greater than the first weight coefficient, and the fourth weight coefficient is less than the second weight coefficient. The fused wheel speed signal is obtained based on the third weighting coefficient, the first weighting coefficient, the fourth weighting coefficient, the second weighting coefficient, the first corrected wheel speed signal, and the second corrected wheel speed signal.

5. The method according to claim 4, characterized in that, After obtaining the fused wheel speed signal based on the third weighting coefficient, the first weighting coefficient, the fourth weighting coefficient, and the second weighting coefficient, as well as the first modified wheel speed signal and the second modified wheel speed signal, the method further includes: Based on preset tools, the brake engagement time and shutdown time of the fused wheel speed signal are extracted; Calculate the attenuation slope and speed fluctuation of the fused wheel speed signal at the corresponding portions of the brake engagement time and the stopping time; When the attenuation slope is greater than the preset slope, or the speed fluctuation is greater than the first fluctuation threshold, the corresponding part is corrected so that the attenuation slope of the corresponding part is less than or equal to the preset slope, and the speed fluctuation is less than or equal to the first fluctuation threshold.

6. The method according to claim 2, characterized in that, The fusion model filters out noise components from the first wheel speed signal to obtain a processed wheel speed signal, including: The first wheel speed signal is decomposed into a wavelet component to obtain a low-frequency component and several thousand high-frequency components. Remove the abnormal noise from the high-frequency components to obtain the effective high-frequency components; The low-frequency component and the effective high-frequency component are reverse-reconstructed to obtain the processed wheel speed signal.

7. The method according to claim 2, characterized in that, The algorithm, based on a preset algorithm, extracts multiple anchor points from all the second collection points, including: Calculate the volatility and average velocity of all second acquisition points in each sliding window of a preset length, wherein the sliding window slides from the starting point on the time axis toward the ending point on the time axis on the second wheel speed signal; Based on the volatility and average velocity, all stable intervals are determined, wherein the volatility of all sliding windows corresponding to the stable interval is less than the second volatility threshold, and the difference in average velocity between two adjacent sliding windows in all sliding windows corresponding to the stable interval is less than the first difference threshold. Extract the second acquisition point located in the middle of each stable interval as the anchor point.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory coupled together, the memory storing a computer program that, when executed by the processor, causes the electronic device to perform the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 7.

10. A program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images