An elevator fault prediction and early warning method and system

Abstract

The present application relates to the technical field of elevator fault prediction, and provides an elevator fault prediction and early warning method and system, wherein the present application synchronously collects elevator operation acoustic signals and traction current, distinguishes uplink and downlink, extracts envelope area according to floor segmentation to represent local energy, pairs uplink and downlink corresponding floor energy to obtain difference to amplify asymmetric faults and offset common mode noise; then, the current amplitude is used to normalize the direction difference, eliminate the load interference, and make the characteristics comparable. Subsequently, the historical calibration direction difference is stored in the database and the slope is fitted to quantify the normal aging drift, the adjacent floor space difference and the drift rate are combined and weighted, the historical mean is used to construct the reference, the threshold is set according to the deviation amount, and the abnormal floor is accurately positioned; finally, the abnormal position time sequence is fitted and extrapolated, the prediction value is greater than the maximum normal value, and early warning is realized to realize early prediction.

Description

Technical Field

[0001] This invention relates to the field of elevator fault prediction technology, and more specifically, to an elevator fault prediction and early warning method and system. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] As an indispensable vertical transportation device in high-rise buildings, the safety and reliability of elevators directly affect the life and property safety of passengers. With the continuous advancement of urbanization in my country, the number of high-rise buildings is increasing daily, and the number of elevators is growing rapidly. Online monitoring and fault early warning of elevator operation status have become crucial for ensuring public safety, reducing maintenance costs, and improving service quality. Currently, elevator operation status monitoring mainly relies on regular manual inspections of the operating parameters of key components such as traction machines, brakes, and door systems, or on setting fixed thresholds based on single sensor signals to trigger over-limit alarms. These methods typically only trigger alarms when a fault has already occurred or the deterioration has progressed to a relatively obvious stage. They are insufficient for sensitively capturing and predicting early, slight deterioration trends in equipment, exhibiting significant detection lag and failing to meet the needs of efficient operation and maintenance management for large-scale elevator clusters.

[0004] In actual operation, factors such as component wear, guide rail deformation, deterioration of lubrication, and installation deviations in the elevator mechanical system often result in differentiated operating characteristics at different operating directions and floors. However, most existing technologies treat elevator operation as a holistic process for general analysis, failing to distinguish between upward and downward operating states, and lacking the ability to precisely locate and analyze specific floors. This makes it easy for local anomalies occurring on specific floors or in specific directions to be masked by the overall average signal, hindering accurate fault location and targeted maintenance. Furthermore, changes in elevator load significantly interfere with operating status signals. The amplitude and frequency domain characteristics of signals collected under different passenger load conditions vary considerably. Existing technologies typically do not incorporate load factors into the signal calibration system, resulting in a lack of a unified comparability benchmark for monitoring data under different operating conditions, thus affecting the accuracy and stability of anomaly detection.

[0005] Meanwhile, existing technologies mostly rely on isolated judgments based on single-run data, failing to fully utilize historical information accumulated from previous runs to build long-term evolution models of equipment status. They cannot effectively distinguish between normal aging drift and sudden anomalies, nor can they quantify and predict the development trend of abnormal locations. This results in delayed early warnings or a high false alarm rate, making it difficult to meet the actual needs of modern intelligent elevator operation and maintenance for early warning and accurate diagnosis. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting and warning of elevator malfunctions, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for predicting and warning of elevator malfunctions, including: S101, in response to the elevator start signal, the microphone installed in the elevator collects the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and collects the downward audio signal and the downward current signal of the traction motor when the elevator is going down. S102, the uplink audio signal and downlink audio signal are segmented according to the floor position to obtain uplink segmented signal and downlink segmented signal respectively. The envelope of the uplink segmented signal and downlink segmented signal is extracted. The envelope area of ​​the extracted corresponding envelope is calculated to obtain the uplink energy sequence and downlink energy sequence respectively. S103, pair the uplink energy sequence and downlink energy sequence one by one according to the same floor position, and calculate the difference between each pair of uplink energy sequence and downlink energy sequence to obtain the direction difference sequence; S104: Extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; use the load calibration value to normalize the direction difference sequence to obtain the calibration direction difference sequence. S105, the calibration direction difference sequence is stored in the historical fingerprint database as the direction acoustic fingerprint of this run. The direction acoustic fingerprint in the historical fingerprint database is linearly fitted according to the order of the runs to obtain the fitting slope, and the fitting slope is used as the direction difference drift rate. S106, calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain the spatial difference sequence; multiply the direction difference drift rate with the spatial difference sequence to obtain the weighted difference sequence; calculate the average value of the weighted difference sequence for each position in each run to obtain the reference weighted sequence; calculate the average deviation of each element in the reference weighted sequence to obtain the fluctuation threshold; mark the floor positions in the weighted difference sequence whose difference with the corresponding position in the reference weighted sequence exceeds the fluctuation threshold as abnormal floor positions; S107: Extract the calibration direction difference of the abnormal floor location from the historical fingerprint database to form the abnormal location time series; fit the abnormal location time series to obtain the trend curve and extrapolate to obtain the predicted direction difference; when the predicted direction difference exceeds the maximum value of the calibration direction difference in the normal stage of the abnormal floor location, generate an early warning signal.

[0007] Furthermore, envelope extraction is performed on the uplink segmented signal and the downlink segmented signal, specifically including: Hilbert transforms are performed on the uplink and downlink segmented signals respectively to obtain analytic signals. The magnitude of the analytic signals is calculated to obtain the instantaneous envelope. The instantaneous envelope is smoothed by sliding median filtering, and the smoothed envelope is used as the corresponding envelope.

[0008] Furthermore, the area of ​​the extracted envelope is calculated, specifically including: The corresponding envelope is divided into intervals along the time axis according to the sampling point interval. The average value of the envelope at the beginning and end of each interval is calculated and multiplied by the time span of the interval to obtain the area of ​​the interval. The area of ​​the envelope is obtained by summing the areas of all intervals.

[0009] Furthermore, the orientation difference sequence is normalized using the load calibration amount, specifically including: The load calibration quantity is divided into calibration sequences with the same number of elements as the orientation difference sequence according to the floor location; each element in the orientation difference sequence is divided by the element at the corresponding position in the calibration sequence to obtain the normalized result, and the normalized result is arranged according to the floor location to form the calibration orientation difference sequence.

[0010] Furthermore, the directional acoustic fingerprints in the historical fingerprint database are linearly fitted according to the order of operation to obtain the fitting slope, specifically including: Extract directional acoustic fingerprints from the historical fingerprint database in chronological order of operation, and calculate the difference in the calibration direction at the same floor position in every two adjacent directional acoustic fingerprints of operation to obtain the adjacent difference value. Directional acoustic fingerprints with adjacent differences greater than the arithmetic mean of all adjacent differences are marked as outliers and removed; the calibration directional differences of the same floor position in the remaining directional acoustic fingerprints are arranged in the order of operation to form a position time sequence. The location time series is fitted with a straight line using the least squares method to obtain the fitting slope of each floor location; the arithmetic mean of the fitting slopes of each floor location is then calculated to obtain the directional drift rate.

[0011] Furthermore, the fluctuation threshold is obtained by calculating the average deviation of each element in the benchmark weighted sequence, specifically including: The median of all elements in the benchmark weighted sequence is taken as the benchmark center value. The absolute value of the difference between each element in the benchmark weighted sequence and the benchmark center value is calculated. The median absolute deviation is obtained by taking the median of all absolute values ​​and is used as the fluctuation threshold.

[0012] Furthermore, a trend curve is obtained by fitting the time series data of the abnormal location, and the predicted direction difference is obtained by extrapolation, specifically including: The time series of abnormal locations were fitted using a weighted moving average method to obtain trend curves; Extend the trend curve along the running time axis to subsequent running times, and read the value of the trend curve corresponding to the subsequent running times as the prediction direction difference.

[0013] Secondly, this application also provides an elevator fault prediction and early warning system, including: The signal acquisition module is used to respond to the elevator start signal. The microphone installed in the elevator acquires the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and acquires the downward audio signal and the downward current signal of the traction motor when the elevator is going down. The energy extraction module is used to segment the uplink and downlink audio signals according to the floor position, obtain uplink segmented signals and downlink segmented signals respectively, extract the envelope of the uplink segmented signals and downlink segmented signals, calculate the envelope area of ​​the extracted corresponding envelopes, and obtain the uplink energy sequence and downlink energy sequence respectively. The directional difference calculation module is used to pair up the uplink energy sequence and downlink energy sequence one by one according to the same floor position, and calculate the difference between each pair of uplink energy sequence and downlink energy sequence to obtain the directional difference sequence; The load calibration module is used to extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; the load calibration value is used to normalize the direction difference sequence to obtain the calibration direction difference sequence. The drift rate calculation module is used to store the calibration direction difference sequence as the direction acoustic fingerprint of this run into the historical fingerprint database, perform linear fitting on the direction acoustic fingerprints in the historical fingerprint database according to the running order to obtain the fitting slope, and use the fitting slope as the direction difference drift rate. The anomaly localization module is used to calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain a spatial difference sequence; multiply the direction difference drift rate by the spatial difference sequence to obtain a weighted difference sequence; average the weighted difference sequence position by position in each run to obtain a reference weighted sequence; calculate the average deviation of each element in the reference weighted sequence to obtain the fluctuation threshold; and mark the floor positions in the weighted difference sequence whose difference from the corresponding position in the reference weighted sequence exceeds the fluctuation threshold as abnormal floor positions. The fault early warning module is used to extract the calibration direction difference of the abnormal floor location from the historical fingerprint database to form the abnormal location time series; the abnormal location time series is fitted to obtain a trend curve and extrapolated to obtain the predicted direction difference; when the predicted direction difference exceeds the maximum value of the calibration direction difference in the normal stage of the abnormal floor location, an early warning signal is generated.

[0014] Thirdly, this application also provides an electronic device, including: Memory, used to store computer programs; A processor is used to implement the method steps as described in the first aspect when executing a computer program.

[0015] Fourthly, this application also provides a readable storage medium on which a computer program is stored, which, when executed by a processor, implements the method steps of the first aspect.

[0016] The beneficial effects of this invention are as follows: This invention simultaneously collects acoustic signals and traction motor current during elevator operation, and records them separately for upward and downward travel, so that the mechanical state in different directions can be presented independently, avoiding the smoothing out of directional differences by overall analysis; then, the acoustic signals are segmented according to the floor position and the envelope area of ​​each segment is extracted to characterize local energy accumulation, decomposing the continuous operation process into spatial distribution features that can be compared point by point, so that local anomalies occurring on specific floors are no longer submerged by the average signal throughout the process.

[0017] Based on this, the energy of the corresponding floors in the up and down directions is paired one by one and the difference is calculated. Since the energy of the up and down directions tends to be symmetrical during normal operation, while the fault often presents an asymmetry, the difference amplifies the fault characteristics and cancels the common-mode noise that is independent of the direction. At the same time, the load calibration quantity is constructed by the current amplitude to normalize the direction difference. Since the current is directly related to the traction torque, it can effectively eliminate the interference of passenger volume fluctuation on the signal amplitude, so that the characteristics under different load conditions have a unified and comparable benchmark.

[0018] Furthermore, the calibration direction differences from each operation are stored in a historical database in chronological order, and the slope of change is fitted to quantify the slow drift caused by normal equipment aging, thus distinguishing between long-term gradual changes and sudden anomalies. Combined with the spatial differences between adjacent floors and weighted by the drift rate, a baseline distribution is constructed using the historical averages, and an adaptive threshold is set using the average deviation. This allows for the precise identification of abnormal floors that deviate from the normal evolution trajectory. Finally, the historical time series of abnormal floors is trend-fitted and extrapolated. When the predicted value exceeds the maximum value of the normal phase at that location, an early warning is triggered, thereby achieving a shift from delayed alarms to early prediction. Attached Figure Description

[0019] Figure 1 A flowchart of an elevator fault prediction and early warning method provided by the present invention; Figure 2 This is a schematic diagram illustrating the calculation of the directional drift rate in this invention; Figure 3 This is a trend fitting and extrapolation prediction diagram of the calibration direction difference of abnormal floor positions in this invention; Figure 4 This is a schematic diagram of an elevator fault prediction and early warning system provided by the present invention; Figure 5 This is a schematic diagram of an electronic device provided by the present invention.

[0020] In the diagram: 201, signal acquisition module; 202, energy extraction module; 203, direction difference calculation module; 204, load calibration module; 205, drift rate calculation module; 206, anomaly location module; 207, fault early warning module; 301, processor; 302, memory. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] like Figure 1 As shown in the figure, an elevator fault prediction and early warning method proposed in this embodiment of the invention includes: S101, in response to the elevator start signal, the microphone installed in the elevator collects the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and collects the downward audio signal and the downward current signal of the traction motor when the elevator is going down. Specifically, during elevator operation, the torque fluctuation of the traction motor is directly physically coupled with the mechanical vibration of the car, the friction of the guide rail, and the change in the tension of the wire rope. The audio signal can capture the propagation characteristics of mechanical vibration in the air without contact, while the current signal reflects the load change of the traction motor in real time. The synchronous acquisition of the two can provide a dual-modal basis of sound and electricity for subsequent fault diagnosis.

[0023] In practical implementation, an omnidirectional microphone can be installed on the top of the car or next to the traction machine, with a sampling frequency of no less than 16kHz to cover the main frequency band of elevator mechanical vibration. Simultaneously, a Hall current sensor is connected to the traction motor power supply circuit to collect the instantaneous value of the motor phase current at a rate no less than one-tenth of the audio sampling frequency. When the elevator control cabinet outputs a start signal, this signal simultaneously triggers the microphone and current sensor to begin collecting data, ensuring that the two signals are strictly aligned on the time axis. The principle behind this is that synchronous triggering can eliminate the time misalignment of acoustic and electrical characteristics, thereby improving the accuracy of subsequent direction difference calculations.

[0024] In response to the elevator start signal, a microphone installed in the elevator collects the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and collects the downward audio signal and the downward current signal of the traction motor when the elevator is going down. Taking the car going from the first floor to the tenth floor as an example, the microphone continuously records the sound signals generated by the guide rail joints, guide shoe friction, and wire rope vibration as the car passes through each floor. The current sensor simultaneously records the current waveform of the traction motor in each stage of acceleration, constant speed, and deceleration. Since the direction of the car's movement relative to the guide rail is opposite when going up and down, there is an inherent difference in the friction direction and force state at the same floor position. This difference is the physical basis for the subsequent direction difference sequence. Collecting the upward and downward signals separately can completely preserve this direction-related acoustic feature.

[0025] Furthermore, during the data acquisition process, the leveling signal from the floor encoder can be used to truncate the audio and current signals, retaining only the effective segments of the car's movement between adjacent floors and eliminating irrelevant noise during door opening / closing and standby phases. This reduces interference in subsequent envelope extraction and energy calculation, improving the stability and repeatability of directional acoustic fingerprints.

[0026] S102, the uplink audio signal and downlink audio signal are segmented according to the floor position to obtain uplink segmented signal and downlink segmented signal respectively. The envelope of the uplink segmented signal and downlink segmented signal is extracted. The envelope area of ​​the extracted corresponding envelope is calculated to obtain the uplink energy sequence and downlink energy sequence respectively.

[0027] Specifically, the core of this step lies in decomposing the continuously running acoustic signal into discrete energy representations corresponding to each floor position. The principle is that as the elevator car passes through each floor interval, the acoustic signals excited by the guide rail joints, guide shoe friction, and wire rope vibration exhibit a location-dependent energy distribution in the time domain. Segmenting by floor position decouples this spatially coupled acoustic signature from the continuous signal, thus providing position-aligned input for subsequent direction difference calculations. This ensures direct comparability of the energy at the same floor position for both upward and downward travel, avoiding time misalignment caused by fluctuations in operating speed.

[0028] In practice, the leveling pulses output by the floor encoder can be used as the segmentation reference. Whenever the car reaches the leveling position of a certain floor, the leveling pulse triggers a segmentation boundary marker. The audio segment between two adjacent leveling pulses corresponds to the segmented signal for that floor interval. The segmentation expression is as follows:

[0029]

[0030] In the formula, For the upward row The first segment of the audio signal for each floor interval Each sample value; For the next downlink The first segment of the audio signal for each floor interval Each sample value; For the uplink continuous audio signal at time The possible values ​​of ; For the downlink continuous audio signal at time The possible values ​​of ; For the car to reach the first position during the upward process The moment of the floor leveling position; For the car to reach the first position during the descent The moment of the floor leveling position; The audio sampling period; The sequence number of the audio sampling point within the segment; This refers to the floor range number. =1,2,..., ; This represents the total number of floor intervals.

[0031] Taking the car's ascent from the first floor to the tenth floor as an example, the upward audio signal is sequentially truncated by the leveling pulse into nine upward segments: from the first floor to the second floor, from the second floor to the third floor, and so on up to the ninth floor to the tenth floor. Similarly, the downward audio signal is truncated into nine downward segments corresponding to the floor intervals. The principle is that the leveling pulse and the physical position of the car are strictly aligned. Using this as a boundary, the upward and downward segmented signals can be precisely aligned in the dimension of floor position, eliminating the segment length difference introduced by the different durations of acceleration and deceleration phases, so that the subsequent comparison of the envelope area is based on the same spatial span.

[0032] Envelope extraction is performed on the uplink and downlink segmented signals, specifically including: performing Hilbert transforms on the uplink and downlink segmented signals respectively to obtain analytic signals; calculating the magnitude of the analytic signals to obtain the instantaneous envelope; smoothing the instantaneous envelope using a sliding median filter; and using the smoothed envelope as the corresponding envelope. The expression for the Hilbert transform is:

[0033] In the formula, For the upward row The analytic signal of a segmented signal is a complex number; For the upward row Real-value sampling of each segment of audio signal; For the upward row The orthogonal components of the segmented signal after Hilbert transform; The imaginary unit, 2 =-1; Pi; The auxiliary sampling point number in the summation of the Hilbert transform. ≠ ; For the upward row The total number of sampling points in each segment; This refers to the audio sample point number; This refers to the floor range number.

[0034] The expression for the instantaneous envelope is:

[0035] In the formula, For the upward row The segmented signal in the first The instantaneous envelope value of each sampling point; For the upward row An analytic signal (complex number) of a segmented signal; For the upward row The first segment of the audio signal for each floor interval Each sample value (analyzing the real part of the signal); For the upward row Hilbert orthogonal components of a segmented signal (imaginary part of an analytic signal). This refers to the audio sample point number; This refers to the floor range number.

[0036] The principle lies in transforming the real-valued audio signal into a complex analytic signal. The real part is the original signal, and the imaginary part is the orthogonal component of the original signal. The magnitude synthesized from these two parts can eliminate carrier phase oscillations while retaining the slowly varying profile of the signal amplitude, i.e., the instantaneous envelope. This process is essentially amplitude demodulation of the acoustic signal, separating the amplitude changes containing mechanical vibration information from the high-frequency carrier, reducing the sampling point dependence of subsequent energy calculations. In specific implementation, a Hilbert transform is performed independently on each uplink and downlink segment signal. The transform length is consistent with the number of sampling points of the segment signal. After obtaining the corresponding analytic signal, the magnitude is calculated point by point to obtain the instantaneous envelope. Considering that the instantaneous envelope may still contain spikes caused by instantaneous impacts or electrical interference, a sliding median filter is used to smooth it. The window length can be taken as an odd number of sampling points corresponding to one-hundredth of the sampling frequency. The principle of median filtering is to replace the center value with the median value after sorting within the window. This can effectively suppress isolated spikes while preserving the overall trend of the envelope, avoiding excessive contributions from individual impact points to the envelope area calculation, and improving the stability of the energy sequence.

[0037] The area of ​​the extracted envelope is calculated by: dividing the envelope into intervals along the time axis according to the sampling point intervals; averaging the envelope values ​​at the beginning and end of each interval and multiplying the average by the time span of the interval to obtain the interval area; and summing the areas of all intervals to obtain the total envelope area. The expression for the envelope area is as follows:

[0038]

[0039] In the formula, For the upward row The envelope area (sound energy characterization value) of the segmented signal in each floor interval. For the next downlink The envelope area (sound energy characterization value) of the segmented signal in each floor interval. For the upward row The envelope of the segmented signal after median filtering and smoothing is at the _th The value of the point; The audio sampling period; For the upward row The total number of sampling points for segmented audio signals in each floor section; For the next downlink The total number of sampling points for segmented audio signals in each floor section; This refers to the audio sample point number; For floor interval numbers; This represents the total number of floor intervals.

[0040] This calculation essentially uses trapezoidal numerical integration to approximate the area under the envelope curve. The principle is that the change in the envelope curve between adjacent sampling points can be approximated as linear. The average height of the interval is represented by the average of the first and last sampling points, and multiplied by the time span to obtain the approximate area of ​​that interval. Summing up the areas of all intervals yields the envelope area of ​​the entire segmented signal. The advantages are low computational cost and sufficient accuracy to characterize the total energy of the segmented signal, while avoiding the direct summation bias caused by differences in sampling rates.

[0041] In practice, the time span is taken as the time interval between adjacent sampling points, i.e., the reciprocal of the audio sampling frequency. The trapezoidal integral described above is performed on each uplink and downlink segment signal to obtain the uplink and downlink envelope areas corresponding to each floor position. Arranging the uplink envelope areas of all floor positions in floor order constitutes the uplink energy sequence, and similarly, arranging the downlink envelope areas in floor order constitutes the downlink energy sequence. Taking the uplink process as an example, if the envelope area of ​​the segment signal from the first to the second floor is calculated as one value, and the envelope area of ​​the segment signal from the second to the third floor is calculated as another value, and so on up to the ninth to the tenth floor, arranging the nine values ​​in order yields an uplink energy sequence of length nine. The downlink energy sequence is obtained similarly. The principle is that the envelope area quantitatively reflects the total energy level of the sound signal within the corresponding floor interval, and the energy level is directly related to mechanical friction and vibration intensity. This transforms the incomparable time-domain waveform into a scalar energy sequence that can be paired position by position, laying the data foundation for the position-by-position difference calculation of uplink and downlink energy in S103.

[0042] S103, pair the uplink energy sequence and downlink energy sequence one by one according to the same floor position, and calculate the difference between each pair of uplink energy sequence and downlink energy sequence to obtain the direction difference sequence.

[0043] Specifically, the upward and downward energy sequences are paired one by one according to the same floor position. The principle is that when the elevator car travels up and down through the same floor interval, the direction of the car's movement relative to the guide rail is opposite. The friction direction between the guide shoe and the guide rail working surface, the force side of the wire rope, and the vibration mode caused by the car's off-center load all have inherent differences. This difference is manifested in the acoustic signal energy as a systematic offset between the upward and downward energy. By pairing and subtracting the upward and downward energy at the same floor position, common noises unrelated to direction (such as ambient background noise and control cabinet fan noise) can be canceled out, while the direction-related acoustic features are highlighted in the form of differences, thus achieving the extraction of direction-sensitive features. Calculating the direction difference sequence compresses the two energy information paths of upward and downward that originally needed to be analyzed separately into a single position-related feature. This reduces the data dimension of subsequent processing and suppresses common-mode interference through differential operations, allowing each element in the sequence to directly represent the degree of asymmetry between the upward and downward acoustic energy at that floor position. This provides a direction-sensitive input basis for load calibration using current signals in S104.

[0044] The expression for the direction difference sequence is as follows:

[0045] In the formula, For the first The difference between upward and downward energy at each floor location (directional difference); For the upward row The envelope area (sound energy characterization value) of the segmented signal in each floor interval. For the next downlink The envelope area (sound energy characterization value) of the segmented signal in each floor interval. For floor interval numbers; This represents the total number of floor intervals.

[0046] S104: Extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; use the load calibration value to normalize the direction difference sequence to obtain the calibration direction difference sequence.

[0047] Specifically, this step aims to establish a quantitative correlation between the directional difference characteristics of the acoustic signal and the actual load state of the traction motor. The principle is that when the elevator car operates between different floors, the output torque of the traction motor inherently differs due to differences in floor height, uneven car load distribution, and uneven distribution of the wire rope's self-weight. Since the amplitude of the motor phase current has an approximately linear relationship with the output torque, the current amplitude can objectively characterize the mechanical load level of the car at that floor. Dividing the directional difference sequence by the load calibration amount at the corresponding location essentially eliminates the scaling effect of load fluctuations on the acoustic energy difference, providing a unified and comparable benchmark for the directional difference characteristics under different floor locations and operating conditions. This improves the stability and cross-condition consistency of the directional acoustic fingerprint.

[0048] In practice, the upward and downward current signals are truncated according to the same floor leveling pulse boundaries as the audio segments, resulting in upward and downward current segments corresponding one-to-one with each floor interval. For each current segment, the effective value is calculated: first, the instantaneous current value at each sampling point within the segment is squared; then, the arithmetic mean of the squared results is taken, and the square root is used as the current amplitude at that floor location. The effective values ​​of the current segments are:

[0049]

[0050] In the formula, For the upward row The effective value (current amplitude) of the current segment in each floor interval; For the next downlink The effective value (current amplitude) of the current segment in each floor interval; For the upward row The first current segment The instantaneous current value at each sampling point; For the next downlink The first current segment The instantaneous current value at each sampling point; For the upward row The total number of sampling points for each current segment; For the next downlink The total number of sampling points for each current segment; This refers to the current sampling point number; For floor interval numbers; This represents the total number of floor intervals.

[0051] The principle is that the effective value can comprehensively reflect the energy level of the current throughout the entire floor range, avoiding amplitude estimation deviations caused by instantaneous spikes or ripples, and directly corresponding to the average torque of the motor in that range. The current amplitudes at each floor position in the upward direction are arranged in floor order to form an upward calibration subsequence, and similarly, the current amplitudes at each floor position in the downward direction are arranged in floor order to form a downward calibration subsequence. The average of the elements at the same floor position in both the upward and downward calibration subsequences is calculated to obtain the load calibration amount for that floor position. The load calibration amounts at all floor positions, arranged in floor order, constitute the calibration sequence. The number of elements in this calibration sequence is strictly consistent with the number of elements in the direction difference sequence. The expression for the load calibration amount is as follows:

[0052] In the formula, For the first Load calibration values ​​at each floor location; For the upward row The effective value (current amplitude) of the current segment in each floor interval; For the next downlink The effective value (current amplitude) of the current segment in each floor interval; For floor interval numbers; This represents the total number of floor intervals.

[0053] The normalization of the orientation difference sequence using load calibration values ​​specifically includes: dividing the load calibration values ​​into calibration sequences with the same number of elements as the orientation difference sequence according to floor location; dividing each element in the orientation difference sequence by the corresponding element in the calibration sequence to obtain the normalized result; and arranging the normalized results according to floor location to form the calibration orientation difference sequence. The normalization expression is as follows:

[0054] In the formula, For the first The calibration direction difference at each floor location (direction difference under unit load); For the first The difference between upward and downward energy at each floor location (directional difference); The load calibration value for the k-th floor location; For floor interval numbers; This represents the total number of floor intervals.

[0055] Taking the car's ascent from the first floor to the tenth floor as an example, if the directional difference sequence length is nine, corresponding to nine floor intervals from the first to the second to the ninth to the tenth floor, then the calibration sequence also contains nine elements, which are the average of the effective values ​​of the upward and downward currents in each interval. Dividing the first element in the directional difference sequence by the first element in the calibration sequence yields the normalized result for the first to second floor interval, and so on until the ninth interval. Arranging the nine normalized results in floor order gives the calibration directional difference sequence. Its beneficial effect is that when a certain floor interval has a high traction motor current and a large sound energy difference due to a heavy car load, the normalization operation can convert the directional difference characteristics of that interval to a unit load level, avoiding the systematic amplification of the directional difference in high-load intervals. This ensures that each element in the calibration directional difference sequence truly reflects the degree of directional asymmetry of mechanical friction and vibration.

[0056] S105, the calibration direction difference sequence is stored in the historical fingerprint database as the direction acoustic fingerprint of this run. The direction acoustic fingerprints in the historical fingerprint database are linearly fitted according to the order of the runs to obtain the fitting slope. The fitting slope is used as the direction difference drift rate.

[0057] Specifically, after each elevator operation and processing via S101 to S104, the resulting calibration direction difference sequence constitutes a directional acoustic fingerprint corresponding to that operation. This fingerprint, indexed by floor location and using calibration direction difference values ​​as elements, quantitatively records the degree of asymmetry in the upward and downward acoustic energy between each floor interval during the operation. It is then sequentially written into the historical fingerprint database according to the operation timestamp. The principle behind this is that the degradation processes of the elevator mechanical system, such as wear, guide rail deformation, and guide shoe aging, have a slow, cumulative characteristic along the time axis. A single operation's fingerprint only reflects a snapshot of the state at a particular moment, while arranging the fingerprints in chronological order allows for the reconstruction of the temporal evolution trajectory of the degradation process, thus providing a longitudinal data foundation for trend identification. By accumulating historical data, instantaneous state observations are transformed into sequence data suitable for time-series analysis, making subsequent drift rate calculations statistically significant rather than relying on the randomness of a single measurement.

[0058] The method involves linearly fitting directional acoustic fingerprints from the historical fingerprint database according to their operational sequence to obtain the fitting slope. Specifically, this includes the following implementation: First, directional acoustic fingerprints are extracted from the historical fingerprint database in chronological order. Then, the difference in calibration direction at the same floor location is calculated between every two adjacent directional acoustic fingerprints to obtain the adjacent difference value. The expression for the adjacent difference value is as follows:

[0059] In the formula, For the first Next and first -1 adjacent differences in calibration direction at the kth floor position during the first run; For the first During the second run The calibration direction difference at each floor location; For the first -1st run The calibration direction difference at each floor location; The sequence numbers are arranged chronologically from earliest to latest. This refers to the floor range number.

[0060] In practical implementation, the sliding window length of the historical fingerprint database can be set. For example, the directional acoustic fingerprints from the most recent 30 runs can be taken, sorted from earliest to latest by run timestamp, and numbered sequentially from the first to the thirtieth. For each floor location, the calibration direction difference of the second run is subtracted from the corresponding value of the first run, the third run is subtracted from the second run, and so on, resulting in 29 adjacent differences for that floor location. The adjacent differences of all floor locations together constitute the adjacent difference set. The principle is that the fingerprint differences of adjacent runs can sensitively reflect the magnitude of changes in state within a short period of time. During normal degradation, the adjacent differences should be within a small and stable range, while fingerprint abrupt changes caused by instantaneous interference, abnormal loads, or occasional sensor malfunctions will cause the corresponding adjacent differences to be significantly larger. By calculating the adjacent differences, potential abnormal fingerprints can be initially located from the normal sequence. The beneficial effect is that it can identify anomalies based on the statistical distribution of the sequence itself without relying on external thresholds, adapting to the individual differences of different elevators.

[0061] Next, directional acoustic fingerprints with adjacent differences greater than the arithmetic mean of all adjacent differences are marked as outliers and removed. Specifically, the arithmetic mean of all elements in the aforementioned set of adjacent differences is first calculated as the discrimination criterion. The expression for the arithmetic mean of adjacent differences is:

[0062] In the formula, The arithmetic mean of all adjacent differences (outlier fingerprint discrimination criterion); For the first Next and first -1st run in the The difference between adjacent calibration directions at each floor location; This represents the total number of floor intervals. Total number of historical runs (within the sliding window); The sequence numbers are arranged chronologically from earliest to latest. This refers to the floor range number.

[0063] Then, each adjacent difference is compared with the mean. If an adjacent difference is greater than the mean, the corresponding acoustic fingerprint from the next run is marked as an outlier and removed from the subsequent fitted data. The principle is that an adjacent difference greater than the arithmetic mean means that the fingerprint from the current run has undergone a jump above the average level compared to the previous one. This jump may be caused by non-degenerative factors rather than the slow degradation of the mechanical system. If retained, it would excessively pull on the linear fitting slope, leading to a distortion in the drift rate estimation. Adaptive removal of outliers is achieved through simple mean comparison, effectively improving the robustness of the fitting results without introducing complex statistical models.

[0064] Subsequently, the calibration direction differences at the same floor location in the remaining directional acoustic fingerprints are arranged in chronological order to form a location time series. In practice, the remaining fingerprints after removing outliers are considered valid samples. For each floor location, the calibration direction difference is extracted from each run, and arranged in chronological order to obtain the location time series for that floor location. The horizontal axis of the location time series represents the run number, and the vertical axis represents the calibration direction difference. The principle is that the change in the calibration direction difference at the same floor location along the chronological order directly reflects the temporal evolution of the degree of asymmetry in the direction of mechanical friction and vibration at that location. After organizing it into a location time series, regression analysis can be used to quantify its rate of change. Decomposing the multidimensional fingerprint data into independent one-dimensional time series for each floor location allows the degradation trend of each location to be evaluated separately, avoiding confusion between differences in degradation rates at different floor locations.

[0065] Then, the least squares method is used to fit a straight line to the position time series, obtaining the fitting slope for each floor position. In practice, for each floor position's position time series, with the run number as the independent variable and the calibration direction difference as the dependent variable, a straight line is fitted using the least squares method. This minimizes the sum of squares of the longitudinal deviations of each data point from this line. The slope of this line represents the average rate of change of the calibration direction difference for that floor position with the number of runs. The expression for the least squares fitting is:

[0066]

[0067]

[0068] In the formula, For the first The least squares fit slope (degradation rate) of the calibration direction difference of each floor location. To remove outliers that are still valid in the location and timing sequence, =1,2,..., ; To effectively run the arithmetic mean of the sequence numbers; To remove outliers, the first The original execution sequence number of a valid running operation; For the first The first effective run in the The calibration direction difference at each floor location; For the first The arithmetic mean of the differences in all valid operating calibration directions at each floor location; This represents the number of valid fingerprints remaining after removing outlier fingerprints. This refers to the floor range number.

[0069] The principle behind this method is that least squares fitting can extract the dominant linear trend from observation data with random fluctuations. The sign of the slope indicates the increase or decrease of the calibration direction difference, respectively, while the absolute value of the slope quantifies the rate of change. Since elevator mechanical degradation typically exhibits a slow, monotonous or near-monotonous change, the linear slope can effectively characterize the degradation rate. A beneficial effect is that it compresses complex temporal variations into a single scalar slope, facilitating subsequent cross-location integration and threshold comparison.

[0070] Finally, the arithmetic mean of the fitted slopes at each floor location is used to obtain the directional drift rate, such as... Figure 2 As shown. In practice, the arithmetic mean of the fitted slopes for all floor locations is calculated, and the resulting average is the directional drift rate for this calculation. This drift rate characterizes the overall degree of drift of the elevator's directional voiceprint fingerprint along the time axis at the current stage. The expression for the directional drift rate is:

[0071] In the formula, The directional drift rate (the overall drift degree of the directional voiceprint fingerprint of the entire elevator). The least-squares fit slope (degradation rate) for calibrating the orientation difference at the k-th floor location. This represents the total number of floor intervals. This refers to the floor range number.

[0072] Taking a ten-story elevator as an example, if outlier fingerprints are removed and twenty-five valid fingerprints remain, the slopes obtained by least-squares fitting at the nine floor locations are nine numerical values. The arithmetic mean of these slopes yields the directional drift rate. As guide rail wear continues, the fitted slopes at each floor location are mostly positive, and the directional drift rate after arithmetic mean is also positive and increases with wear, thus providing a quantitative trend basis for anomaly detection in S106. The principle is that the arithmetic mean can synthesize the local degradation rate at each floor location, reflecting the overall degradation trend of the elevator and avoiding misjudgment of the overall state due to accidental fluctuations at a single floor location.

[0073] S106, calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain the spatial difference sequence; multiply the direction difference drift rate with the spatial difference sequence to obtain the weighted difference sequence; calculate the average value of the weighted difference sequence for each position in each run to obtain the reference weighted sequence; calculate the average deviation of each element in the reference weighted sequence to obtain the fluctuation threshold; mark the floor positions in the weighted difference sequence whose difference with the corresponding position in the reference weighted sequence exceeds the fluctuation threshold as abnormal floor positions.

[0074] Specifically, this step aims to identify local degradation locations from a spatial dimension. The principle is that the calibration direction difference sequence has eliminated the influence of load scale. The difference between adjacent floor positions can highlight abrupt changes in the direction asymmetry of a certain floor interval relative to its adjacent intervals. Under normal mechanical conditions, the direction difference between adjacent floors should have a smooth transition. Abrupt changes often correspond to loose guide rail joints, localized wear of guide shoes, or abnormal tension in the wire rope. In practice, the calibration direction difference sequence is arranged in floor order. The absolute value of each subsequent element is subtracted from the previous element, resulting in a sequence shorter than the original sequence, which is the spatial difference sequence. Taking a ten-story elevator as an example, nine floor intervals correspond to eight spatial difference values. If the calibration direction difference between the third and fourth floors is significantly higher than that between the second and third floors and between the fourth and fifth floors, then the spatial difference value corresponding to that location is too large. Multiplying the direction difference drift rate by the spatial difference sequence yields a weighted difference sequence. The principle is that the drift rate characterizes the overall degradation rate; using it as a weight allows the spatial difference to achieve higher sensitivity during the accelerated degradation phase, avoiding false alarms due to small fluctuations during the stable period. The baseline weighted sequence is obtained by averaging the weighted difference sequence at each location from each run. The principle behind this is that the average value from multiple runs can suppress single, accidental disturbances, forming a normal baseline for each floor location. The expression for the weighted difference sequence is as follows:

[0075]

[0076] In the formula, For the first The and the first +1 spatial difference value of calibration direction difference between floor locations; For the first The calibration direction difference at each floor location (direction difference under unit load); For the first +1 floor location calibration direction difference (direction difference under unit load); This is the [number]th run The weighted difference values ​​for each position; The directional drift rate (the overall drift degree of the directional voiceprint fingerprint of the entire elevator); For floor interval numbers; This represents the total number of floor intervals.

[0077] The mean value is used to obtain the baseline weighted sequence:

[0078] In the formula, For the first The baseline weighted value for each position (the average of the weighted differences from each run). For the first The second run in The weighted difference values ​​for each position; Total number of historical runs (within the sliding window); The sequence numbers are arranged chronologically from earliest to latest. This refers to the floor range number.

[0079] The fluctuation threshold is obtained by calculating the average deviation of each element in the benchmark weighted sequence. Specifically, this involves: calculating the median of all elements in the benchmark weighted sequence as the benchmark center value; calculating the absolute value of the difference between each element in the benchmark weighted sequence and the benchmark center value; and calculating the median absolute deviation of all absolute values ​​as the median. The median and median absolute deviation are used as the fluctuation threshold. The principle behind using the median and median absolute deviation is that they are insensitive to extreme values, robustly representing the dispersion of the benchmark sequence and preventing individual abnormal positions from raising the threshold and masking other anomalies. Floors in the weighted difference sequence whose difference from the corresponding position in the benchmark weighted sequence exceeds the fluctuation threshold are marked as abnormal floor positions. For example, if the weighted difference value between the third and fourth floors in a certain operation, minus the corresponding value in the benchmark weighted sequence, is greater than the fluctuation threshold, then that position is marked as an abnormal floor position. This achieves joint discrimination of spatial positioning and trend weighting, locating faulty floors while also considering the overall degradation stage.

[0080] S107: Extract the calibration direction difference of the abnormal floor location from the historical fingerprint database to form the abnormal location time series; fit the abnormal location time series to obtain the trend curve and extrapolate to obtain the predicted direction difference; when the predicted direction difference exceeds the maximum value of the calibration direction difference in the normal stage of the abnormal floor location, generate an early warning signal.

[0081] Specifically, based on the abnormal floor locations marked in S106, the calibration direction difference values ​​corresponding to each run are retrieved from the historical fingerprint database in chronological order, using these floor locations as indexes. The principle is that the evolution trajectory of the calibration direction difference at the abnormal location along the time axis directly reflects the mechanical degradation process at that location. Organizing discrete historical observations into time-series data provides input for trend prediction. Taking a ten-story elevator as an example, if S106 marks the section between the third and fourth floors as abnormal floor locations, the calibration direction difference values ​​for this section in the most recent thirty runs are extracted from the historical fingerprint database and arranged in ascending order of run number to obtain a time series of abnormal locations with a length of thirty.

[0082] Then, a trend curve is obtained by fitting the time series of the abnormal location, and the predicted direction difference is obtained by extrapolation. Figure 3 As shown, it specifically includes: A weighted moving average method is used to fit the time series data at abnormal locations to obtain a trend curve. In practice, a sliding window length is set, for example, five runs. The calibration direction differences of each run within the window are assigned weights that increase with the run number, with observations closer to the current time having greater weights. The weighted sum is then used to obtain the trend value at the current time. The trend curve is generated by the window sliding sequentially. The principle is that recent data better reflects the current degradation rate, and weighting suppresses the lagging influence of older data on trend judgment, while smoothing random fluctuations, ensuring that the trend curve faithfully represents the dominant direction of degradation. The expression for the trend curve is:

[0083] In the formula, For the first The weighted moving average trend value at each running time; The first in the weighted moving average The weights of each data point are taken. ; For the first The calibration direction difference of abnormal floor locations during the second operation; The length of the weighted moving average sliding window; This represents the number of lag steps of the data within the weighted moving average window from the current time. =0,1,..., -1; The sequence numbers are arranged chronologically from earliest to latest. Total number of historical runs (within the sliding window); The floor number is the location number that is marked as abnormal.

[0084] The trend curve is extended along the time axis to subsequent running times, and the value of the trend curve at each subsequent running time is used as the predicted direction difference. In practice, based on the most recent trend value and its rate of change, the predicted direction difference for the third subsequent running time is calculated using a linear extrapolation method. The principle is that mechanical degradation is approximately linear within a short time window, and extrapolation can predict the future state at that position in advance. The expression is as follows:

[0085] In the formula, The difference in predicted direction obtained by extrapolation; For the most recent (the) The weighted moving average trend value at each running time (times); For the previous time (the first) The weighted moving average trend value at each of the -1st running times; The extrapolation step number (e.g., predicting the 3rd subsequent run). =3); This represents the total number of historical runs (within the sliding window).

[0086] A warning signal is generated when the predicted direction difference exceeds the maximum value of the calibrated direction difference during normal operation at the abnormal floor location. Specifically, the maximum value of the calibrated direction difference during the ten consecutive normal operations prior to the abnormal floor location being marked as abnormal is taken as the warning threshold. If the predicted direction difference exceeds this threshold, a warning signal is output to the elevator control cabinet. Its advantage lies in providing early warning before a fault actually occurs, allowing maintenance personnel time to handle the situation and preventing operational abnormalities caused by the continuous deterioration of guide rail wear or guide shoe aging.

[0087] like Figure 4 As shown, based on the same inventive concept, this embodiment provides an elevator fault prediction and early warning system, including: The signal acquisition module 201 is used to respond to the elevator start signal. The microphone installed in the elevator acquires the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and acquires the downward audio signal and the downward current signal of the traction motor when the elevator is going down. The energy extraction module 202 is used to segment the uplink audio signal and the downlink audio signal according to the floor position to obtain the uplink segment signal and the downlink segment signal respectively, extract the envelope of the uplink segment signal and the downlink segment signal, calculate the envelope area of ​​the extracted corresponding envelope, and obtain the uplink energy sequence and the downlink energy sequence respectively. The directional difference calculation module 203 is used to pair up the uplink energy sequence and downlink energy sequence one by one according to the same floor position, and calculate the difference between each pair of uplink energy sequence and downlink energy sequence to obtain the directional difference sequence; The load calibration module 204 is used to extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; the load calibration value is used to normalize the direction difference sequence to obtain the calibration direction difference sequence. The drift rate calculation module 205 is used to store the calibration direction difference sequence as the direction acoustic fingerprint of this run into the historical fingerprint database, perform linear fitting on the direction acoustic fingerprint in the historical fingerprint database according to the running order to obtain the fitting slope, and use the fitting slope as the direction difference drift rate. The anomaly location module 206 is used to calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain a spatial difference sequence; multiply the direction difference drift rate with the spatial difference sequence to obtain a weighted difference sequence; calculate the average value of the weighted difference sequence for each position in each run to obtain a reference weighted sequence; calculate the average deviation of each element in the reference weighted sequence to obtain a fluctuation threshold; and mark the floor positions in the weighted difference sequence whose difference with the corresponding position in the reference weighted sequence exceeds the fluctuation threshold as abnormal floor positions. The fault early warning module 207 is used to extract the calibration direction difference of the abnormal floor position in each operation from the historical fingerprint database to form the abnormal position time series; to fit the abnormal position time series to obtain the trend curve and extrapolate to obtain the predicted direction difference; when the predicted direction difference exceeds the maximum value of the calibration direction difference in the normal stage of the abnormal floor position, an early warning signal is generated.

[0088] like Figure 5 As shown, based on the same inventive concept, this embodiment provides an electronic device, including: Memory 302 is used to store computer programs; Processor 301 is used to implement the method steps as described in the first aspect when executing a computer program.

[0089] Based on the same inventive concept, this embodiment provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the method steps as described in the first aspect.

[0090] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0091] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting and warning elevator malfunctions, characterized in that, include: S101, in response to the elevator start signal, the microphone installed in the elevator collects the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and collects the downward audio signal and the downward current signal of the traction motor when the elevator is going down. S102, the uplink audio signal and the downlink audio signal are segmented according to the floor position to obtain uplink segmented signals and downlink segmented signals respectively. The envelopes of the uplink segmented signals and the downlink segmented signals are extracted. The envelope area of ​​the extracted corresponding envelopes is calculated to obtain the uplink energy sequence and the downlink energy sequence respectively. S103, the uplink energy sequence and the downlink energy sequence are paired one by one according to the same floor position, and the difference between each pair of uplink energy sequence and downlink energy sequence is calculated to obtain the direction difference sequence; S104, extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; use the load calibration value to normalize the direction difference sequence to obtain the calibration direction difference sequence; S105, the calibration direction difference sequence is stored in the historical fingerprint database as the direction acoustic fingerprint of this operation, and the direction acoustic fingerprint in the historical fingerprint database is linearly fitted according to the order of operation to obtain the fitting slope, and the fitting slope is used as the direction difference drift rate. S106, Calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain the spatial difference sequence; The directional drift rate is multiplied by the spatial difference sequence to obtain a weighted difference sequence. The average value of the weighted difference sequence for each position in each run is calculated to obtain a reference weighted sequence. The average deviation of each element in the reference weighted sequence is calculated to obtain a fluctuation threshold. Floor positions in the weighted difference sequence whose difference from the corresponding position in the reference weighted sequence exceeds the fluctuation threshold are marked as abnormal floor positions. S107, Extract the calibration direction difference of the abnormal floor location from the historical fingerprint database to form the abnormal location time sequence. A trend curve is obtained by time-series fitting of the abnormal location and the predicted direction difference is obtained by extrapolation. When the predicted direction difference exceeds the maximum value of the calibrated direction difference during the normal phase of the abnormal floor location, an early warning signal is generated.

2. The elevator fault prediction and early warning method according to claim 1, characterized in that, The envelope extraction of the uplink segmented signal and the downlink segmented signal specifically includes: The uplink segmented signal and the downlink segmented signal are subjected to Hilbert transform respectively to obtain analytic signals; the magnitude of the analytic signals is obtained to obtain the instantaneous envelope; the instantaneous envelope is smoothed by sliding median filtering; and the smoothed envelope is used as the corresponding envelope.

3. The elevator fault prediction and early warning method according to claim 2, characterized in that, The step of calculating the envelope area of ​​the extracted corresponding envelope specifically includes: The corresponding envelope is divided into intervals along the time axis according to the sampling point interval. The average value of the envelope at the beginning and end of each interval is calculated and multiplied by the time span of the interval to obtain the area of ​​the interval. The area of ​​the envelope is obtained by summing the areas of all the intervals.

4. The elevator fault prediction and early warning method according to claim 1, characterized in that, The normalization of the orientation difference sequence using the load calibration amount specifically includes: The load calibration amount is divided into calibration sequences with the same number of elements as the direction difference sequence according to the floor position; each element in the direction difference sequence is divided by the element at the corresponding position in the calibration sequence to obtain a normalized result, and the normalized result is arranged according to the floor position to form the calibration direction difference sequence.

5. The elevator fault prediction and early warning method according to claim 1, characterized in that, The process of linearly fitting the directional acoustic fingerprints in the historical fingerprint database according to the order of operation to obtain the fitting slope specifically includes: The directional acoustic fingerprints are extracted from the historical fingerprint database in chronological order of operation. The difference between the calibration direction difference of the same floor position in every two adjacent directional acoustic fingerprints is calculated to obtain the adjacent difference value. The directional acoustic fingerprints whose adjacent differences are greater than the arithmetic mean of all adjacent differences are marked as outliers and removed; the calibration directional differences of the same floor position in the remaining directional acoustic fingerprints are arranged in the order of operation to form a position time sequence. The location time series is fitted with a straight line using the least squares method to obtain the fitting slope of each floor location; the arithmetic mean of the fitting slopes of each floor location is then calculated to obtain the directional drift rate.

6. The elevator fault prediction and early warning method according to claim 1, characterized in that, The process of obtaining the fluctuation threshold by calculating the average deviation of each element in the benchmark weighted sequence specifically includes: The median of all elements in the benchmark weighted sequence is taken as the benchmark center value. The absolute value of the difference between each element in the benchmark weighted sequence and the benchmark center value is calculated. The median absolute deviation is obtained by taking the median of all the absolute values ​​and is used as the fluctuation threshold.

7. The elevator fault prediction and early warning method according to claim 1, characterized in that, The process of fitting a trend curve to the time series of the abnormal location and extrapolating it to obtain the predicted direction difference specifically includes: The time series of the abnormal locations were fitted using a weighted moving average method to obtain a trend curve; The trend curve is extended along the running time axis to subsequent running times, and the value of the trend curve corresponding to the subsequent running times is read as the prediction direction difference.

8. An elevator fault prediction and early warning system, based on the elevator fault prediction and early warning method according to claim 1, characterized in that, include: The signal acquisition module is used to respond to the elevator start signal. The microphone installed in the elevator acquires the upward audio signal and the upward current signal of the traction motor when the elevator is going up, and acquires the downward audio signal and the downward current signal of the traction motor when the elevator is going down. The energy extraction module is used to segment the uplink audio signal and the downlink audio signal according to the floor position to obtain uplink segmented signals and downlink segmented signals respectively, extract the envelope of the uplink segmented signals and the downlink segmented signals, calculate the envelope area of ​​the extracted corresponding envelopes, and obtain the uplink energy sequence and the downlink energy sequence respectively. The direction difference calculation module is used to pair the uplink energy sequence and the downlink energy sequence one by one according to the same floor position, and calculate the difference between each pair of uplink energy sequence and downlink energy sequence to obtain the direction difference sequence; The load calibration module is used to extract the amplitudes of the uplink current signal and the downlink current signal respectively to form the load calibration value; and to normalize the direction difference sequence using the load calibration value to obtain the calibration direction difference sequence. The drift rate calculation module is used to store the calibration direction difference sequence as the directional acoustic fingerprint of the current run into the historical fingerprint database, perform linear fitting on the directional acoustic fingerprint in the historical fingerprint database according to the running order to obtain the fitting slope, and use the fitting slope as the direction difference drift rate. An anomaly localization module is used to calculate the absolute value of the difference between adjacent floor positions in the calibration direction difference sequence to obtain a spatial difference sequence; multiply the direction difference drift rate by the spatial difference sequence to obtain a weighted difference sequence; calculate the average value of the weighted difference sequence for each position in each run to obtain a reference weighted sequence; calculate the average deviation of each element in the reference weighted sequence to obtain a fluctuation threshold; and mark the floor positions in the weighted difference sequence whose difference from the corresponding position in the reference weighted sequence exceeds the fluctuation threshold as abnormal floor positions. The fault early warning module is used to extract the calibration direction difference of the abnormal floor location in each operation from the historical fingerprint database to form the abnormal location time sequence. A trend curve is obtained by time-series fitting of the abnormal location and the predicted direction difference is obtained by extrapolation. When the predicted direction difference exceeds the maximum value of the calibrated direction difference during the normal phase of the abnormal floor location, an early warning signal is generated.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the elevator fault prediction and early warning method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements an elevator fault prediction and early warning method as described in any one of claims 1 to 7.

Images