Fault early warning method for gearless traction machine based on multi-source data fusion

Abstract

The present application relates to the field of traction machine monitoring technology, and more particularly to a gearless traction machine fault early warning method based on multi-source data fusion. The method comprises: obtaining a braking current sequence, a driving voltage sequence and a micro switch state sequence and extracting a measured action time; using a dynamic time warping algorithm to calculate the waveform distortion index between the braking current sequence and the preset standard fingerprint; calculating the voltage correction factor, the resistance correction factor, and correcting the measured action time to obtain the normalized action time; inputting the waveform distortion index and the normalized action time into a local outlier factor algorithm to calculate the equipment health degree score, and combining the logical consistency to calculate the comprehensive risk early warning index. That is, the scheme of the present application considers mechanical form and electrical physical characteristics when evaluating the state of the brake by fusing multi-source data, thereby achieving more accurate fault identification.

Description

Technical Field

[0001] This invention relates to the field of traction machine monitoring technology. More specifically, this invention relates to a fault early warning method for gearless traction machines based on multi-source data fusion. Background Technology

[0002] Permanent magnet synchronous gearless traction machines are the core power equipment of modern elevators, and the electromagnetic brake, as a critical safety component, is directly related to the safe operation of the elevator. When the elevator stops, the electromagnetic brake uses spring force to hold the brake wheel; during operation, energizing a coil generates electromagnetic force to attract the armature, thus releasing the brake. Its reliable operation is the last line of defense against elevator slippage, overshooting, or bottoming out; therefore, real-time status monitoring is crucial. Because the brake's operating environment involves high-frequency mechanical reciprocating motion and complex electrical fluctuations, its condition deterioration is often a hidden, gradual process. Once a failure occurs, the consequences are extremely serious. Therefore, real-time acquisition of its operating parameters and effective fault warnings are paramount to ensuring the safe operation of elevators.

[0003] To monitor the status of the brake, current technologies primarily rely on microswitch feedback and static current threshold detection. Microswitches detect the displacement of the armature, providing a 0 / 1 status signal ("in position" or "out of position") to determine if the action has been completed. Static current threshold detection, on the other hand, sets fixed upper and lower current limits in the electrical circuit and monitors the current in real time to determine if the equipment is malfunctioning. These methods attempt to gain a preliminary understanding of the brake's basic operating status through simple physical signal feedback and quantitative comparison, enabling timely alarm signals in the event of significant electrical or mechanical failures, thereby improving the safety of the elevator system.

[0004] However, the aforementioned existing technologies have the following significant drawbacks in practical applications. First, they cannot eliminate interference from voltage fluctuations, leading to a high false alarm rate. Elevator power supply voltage fluctuates significantly due to building load; according to physical laws, when the voltage decreases, the coil current rise rate slows down, and the magnetic force establishment is delayed, resulting in a longer brake release time. Because existing technologies cannot distinguish between the physical characteristic delay caused by voltage drops and actual mechanical faults, false alarms are highly likely to occur during peak electricity consumption periods. Second, they cannot detect early mechanical lag, leading to a high false alarm rate. Rusting or lubrication drying of the brake pin is a gradual process. Although the microswitch can operate in the early stages of a fault, existing technologies ignore the mechanical motion characteristics inherent in the current establishment process, failing to reflect subtle changes in the waveform. This makes it difficult to issue a warning before the component completely jams, thus affecting the timeliness of fault diagnosis. Summary of the Invention

[0005] The purpose of this invention is to propose a fault early warning method for gearless traction machines based on multi-source data fusion, in order to solve the problems of high false alarm rate caused by the inability to eliminate voltage fluctuation interference and high missed alarm rate caused by the inability to detect early mechanical hysteresis in the prior art; to this end, the present invention provides a solution in one aspect.

[0006] The fault early warning method for gearless traction machines based on multi-source data fusion provided by this invention includes:

[0007] The braking current sequence, driving voltage sequence, and microswitch state sequence during the braking process of a gearless traction machine are obtained, and the measured action time is extracted. A dynamic time warping algorithm is used to calculate the waveform distortion index between the braking current sequence and a preset standard fingerprint to evaluate the morphological characteristics of the mechanical motion. A normalized action response time model is constructed, and a voltage correction factor is calculated based on the driving voltage sequence. A resistance correction factor is calculated based on the braking current sequence. The measured action time is corrected using the voltage correction factor and the resistance correction factor to obtain the normalized action time. The waveform distortion index and the normalized action time are input into a local outlier factor algorithm to obtain an outlier factor. Based on the outlier factor, a device health score is calculated, and a comprehensive risk warning index is calculated by combining the logical consistency between the microswitch state sequence and the braking current sequence to achieve fault warning for the gearless traction machine.

[0008] Thus, by fusing multi-source data, the present invention can simultaneously consider mechanical morphological characteristics and electrophysical characteristics when evaluating the brake status, thereby achieving more accurate fault identification.

[0009] Preferably, the normalized action time satisfies the following relationship:

[0010]

[0011] In the formula, This represents the corrected normalized time; Indicates the actual measured action time; This is the measured average voltage; This is the rated voltage of the brake; This indicates the measured peak current. Indicates the reference current value; , This serves as a safety base to prevent the denominator from being zero. , The sensitivity index is set based on the characteristics of the electromagnetic coil.

[0012] Thus, through this normalization model, a compensation mechanism incorporating both voltage and current factors is introduced for the first time. This mechanism can accurately identify that the extended operating time caused by low voltage or coil heat is a physical phenomenon rather than a mechanical fault, thereby correcting the operating time back to the normal value and effectively eliminating false alarms caused by voltage fluctuations.

[0013] Preferably, the equipment health score satisfies the following relationship:

[0014]

[0015] In the formula, This indicates the device's health score; Indicates the outlier factor; Indicates the bias constant; Indicates the aging coefficient; This indicates the current cumulative number of runs; Indicates the design life.

[0016] In this way, by establishing a health rating model, it is no longer limited to simple alarms that exceed the standard, but provides a visualized health score, enabling maintenance personnel to observe the decline process of equipment from healthy to sub-healthy, which facilitates preventive maintenance before complete failure.

[0017] Preferably, the comprehensive risk warning index satisfies the following relationship:

[0018]

[0019] In the formula, This represents the comprehensive risk warning index; This indicates the device's health score; Indicates the conflict gain coefficient; Indicates the measured state of the micro switch; This indicates the theoretically expected state inferred from the characteristics of the braking current sequence.

[0020] In this way, by combining data trend risk and logical state risk through a comprehensive risk warning index, it can provide low-risk warnings in the early stages of wear and tear, and can also instantly amplify the risk value by using the conflict gain coefficient when a logical conflict (such as jamming) is detected, triggering an emergency stop to ensure the absolute safety of the elevator.

[0021] Preferably, the step of calculating the waveform distortion index between the braking current sequence and the preset standard fingerprint using the dynamic time warping algorithm includes: recording multiple normal operating current waveforms during the elevator commissioning phase and generating an average waveform as the preset standard fingerprint; calculating the sum of the Euclidean distances between the currently acquired braking current sequence and the preset standard fingerprint after alignment, and using the sum as the waveform distortion index.

[0022] Thus, by combining the flexibility of the DTW algorithm with the stability of the average fingerprint, a high-precision, high-fault-tolerant, and easily quantifiable elevator health monitoring method is achieved. Its core value lies in reducing the false alarm rate and improving the ability to perceive potential hazards.

[0023] Preferably, the method for obtaining the measured action time includes: listening to the release command of the elevator main controller and recording the time when the release command is issued as the start time; monitoring the state sequence of the micro switch and recording the time when the micro switch state changes from closed to open as the reversal time; calculating the difference between the reversal time and the start time to obtain the measured action time.

[0024] Preferably, the measured average voltage is the arithmetic mean of the driving voltage sequence from the start time to the flip time; the peak current is the maximum value or steady-state holding current value reached by the braking current sequence within the acquisition window.

[0025] Preferably, the method for determining the theoretically required state includes: analyzing the braking current sequence to identify whether there is a current inflection point characterizing armature engagement; if the current inflection point is identified, the theoretically required state is determined to be an engagement state; if the current inflection point is not identified, the theoretically required state is determined to be a release state.

[0026] Preferably, the method further includes: comparing the comprehensive risk warning index with a preset alarm threshold; and performing a shutdown protection operation when the comprehensive risk warning index exceeds the preset alarm threshold.

[0027] Preferably, the sampling frequency is set to 5kHz and the acquisition window duration is set to 400 milliseconds.

[0028] The beneficial effects of this invention are as follows: It proposes a fusion early warning architecture based on physical model decoupling and multidimensional data anomaly detection. First, it uses a dynamic time warping algorithm to capture microsecond-level mechanical transient distortions, solving the problem that traditional Euclidean distance cannot handle waveform time shifts. Second, it constructs a voltage-temperature dual decoupled time compensation model, using physical laws to restore the true mechanical performance. Finally, it uses a local outlier factor algorithm to divide the health domain and fault domain in the multidimensional feature space. This progressive processing logic not only achieves keen perception of early mechanical hysteresis but also fundamentally solves the industry problem of voltage fluctuation interference, achieving the monitoring goal of zero false alarms and zero missed alarms. Attached Figure Description

[0029] Figure 1 The flowchart illustrating the steps of the gearless traction machine fault early warning method based on multi-source data fusion in this embodiment is shown in the schematic diagram.

[0030] Figure 2This diagram illustrates a comparison of brake response time data between the prior art and the present invention under voltage fluctuation conditions.

[0031] Figure 3 The diagram illustrates the differences in the waveform characteristics of the brake coil current.

[0032] Figure 4 The diagram illustrates the state of data analysis after multi-source feature fusion. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0034] like Figure 1 As shown, the gearless traction machine fault early warning method based on multi-source data fusion in this embodiment includes the following steps:

[0035] Step S1: Obtain the braking current sequence, driving voltage sequence, and micro switch state sequence during the braking process of the gearless traction mechanism, and extract the measured action time.

[0036] To ensure data accuracy and data source reliability, this embodiment deploys high-precision sensors at the brake power output terminal of the traction machine control cabinet. For example, a Hall current sensor is used to collect current, and a voltage divider resistor network is used in conjunction with an AD converter to collect voltage.

[0037] For example, a sampling frequency of 5 kHz (i.e., acquiring one data point every 0.2 milliseconds) can be set. This frequency is sufficient to capture microsecond-level mechanical transient changes during brake operation.

[0038] In an optional embodiment, the measured action time is obtained by means of:

[0039] The system monitors the elevator main controller for brake release commands in real time. The moment the brake release command is issued is recorded as the start time. Simultaneously, the signal level of the micro switch is continuously monitored. When the signal changes from high level (closed) to low level (open), the moment is marked as the toggle moment. Calculate the difference between the flipping time and the starting time to obtain the measured action time. .

[0040] if, It is 0ms. If it is 220ms, then the actual action time is... It is 220ms. This value directly reflects the physical time it takes for the brake to go from being energized to the mechanical contacts actuating.

[0041] The normal operating cycle of most brakes is typically around 200ms. In this embodiment, the acquisition window duration is... The time interval was set to 400 milliseconds, which covers the normal operating cycle of most brakes. During this period, the braking current sequence was recorded synchronously. Driving voltage sequence and micro switch state sequence .

[0042] In this way, by acquiring high-frequency data and extracting effective segments, we can obtain raw data containing rich mechanical motion information and electrical characteristics, which provides a solid data foundation for subsequent interference removal and fault identification.

[0043] Step S2: Calculate the waveform distortion index between the braking current sequence and the preset standard fingerprint using the dynamic time warping algorithm to evaluate the morphological characteristics of the mechanical motion.

[0044] When the brake armature moves under the influence of magnetic force, it cuts magnetic field lines, generating a back electromotive force (EMF). This causes a characteristic inflection point (concave or flattened) in the rising phase of the current waveform. If the mechanical damping increases due to rust or jamming, the armature movement slows down, the back EMF weakens, and this causes the inflection point to flatten and shift later on the time axis. Because mechanical aging causes an overall waveform shift, traditional point-to-point Euclidean distance calculations become invalid due to misalignment.

[0045] Therefore, this step uses the Dynamic Time Warping (DTW) algorithm to calculate the waveform distortion index between the braking current sequence and the preset standard fingerprint.

[0046] Specifically, firstly, during the elevator commissioning phase, the current waveforms of 50 normal operating cycles are recorded, and the average waveform is generated as a preset standard fingerprint. In real-time monitoring, the currently acquired braking current sequence and the preset standard fingerprint are input into the DTW algorithm to calculate the sum of the Euclidean distances between the currently acquired braking current sequence and the preset standard fingerprint after alignment. This sum is used as the waveform distortion index.

[0047] The DTW algorithm uses dynamic programming to find the shortest path between two sequences, allowing for non-linear distortion on the time axis (i.e., automatic alignment of feature points). The calculated optimal path distance is the waveform distortion index. .

[0048] For example, the inflection point of the preset standard fingerprint is at 100ms, while the inflection point of the current braking current sequence is at 130ms. If Euclidean distance is used, the corresponding point error will be huge; while the DTW algorithm will align the two inflection points and compare the shape differences.

[0049] like This indicates that the mechanical state is perfect; if If the waveform exceeds a preset threshold (e.g., 0.5), it indicates that the waveform shape has undergone substantial distortion, suggesting abnormal mechanical damping.

[0050] Thus, the DTW algorithm effectively overcomes the interference of time translation on similarity calculation and can keenly capture early mechanical hysteresis features that microswitches cannot detect.

[0051] Step S3: Construct a normalized action response time model, calculate a voltage correction factor based on the driving voltage sequence, calculate a resistance correction factor based on the braking current sequence, and use the voltage correction factor and the resistance correction factor to correct the measured action time to obtain the normalized action time.

[0052] Specifically, in order to solve the false alarm problem caused by "voltage fluctuations", this step constructs a physical model to decouple and correct the measured action time.

[0053] Normalized action time satisfies the following relationship:

[0054]

[0055] In the formula, This represents the corrected normalized time, indicating pure mechanical performance, and is expressed in milliseconds (ms). The measured action time obtained in step S1 is in milliseconds (ms). This represents the measured average voltage, in units of V; The rated voltage of the brake, in volts (V). It is a constant, with a value of, for example, 110V; The measured peak current is the maximum value or steady-state holding current value reached by the braking current sequence within the acquisition window, reflecting the coil resistance state (temperature influence), and the unit is A; This is a reference current value, in amperes (A). As a constant, the rated current under cold conditions is taken. In this embodiment, 2.0A can be selected; , For safety, we use 1.0 and 0.1 respectively to prevent the denominator from being 0; , The sensitivity index is set based on the characteristics of the electromagnetic coil, and is set to 1.5 and 1.0 respectively.

[0056] Wherein, the measured average voltage is the arithmetic mean of the driving voltage sequence from the start time to the flip time; the peak current is the maximum value or steady-state holding current value reached by the braking current sequence within the acquisition window.

[0057] For example, during peak electricity consumption in buildings, when voltage drops:

[0058] Measured average voltage The voltage drops to 90V (below the rated 110V). As a result, the magnetic force builds up more slowly, and the measured operating time... The time was extended to 250ms (normally it should be 200ms). At this point, the coil temperature was normal, and the measured peak current was [not specified]. It is 2.0A.

[0059] The voltage correction factor is calculated using the above normalized action time formula. Resistance correction factor Corrected normalized time .

[0060] The corrected normalized time of 185ms is less than 200ms, which is within the normal range. The system successfully identified that although the measured time was longer, this was a physical phenomenon caused by low voltage, and the machinery itself was healthy, therefore no alarm was triggered. Conversely, if it were a mechanical jam, the voltage correction factor would be 1, the normalized time would still be 250ms, and the system would correctly trigger an alarm.

[0061] Thus, by decoupling voltage and temperature, this solution can restore the true mechanical response time, fundamentally eliminating false alarms caused by grid voltage fluctuations.

[0062] Step S4: Input the waveform distortion index and the normalized action time into the local outlier factor algorithm to obtain the outlier factor, calculate the equipment health score based on the outlier factor, and calculate the comprehensive risk warning index by combining the logical consistency of the micro-switch state sequence and the braking current sequence to realize the fault warning of the gearless traction machine.

[0063] Specifically, the LOF algorithm is first used to process the waveform distortion index. and the corrected normalized time The two-dimensional feature vector is composed of the outlier factor. Under normal circumstances Abnormal situations .

[0064] Next, calculate the equipment health score. The equipment health score satisfies the expression:

[0065]

[0066] In the formula, This indicates the device's health score; Indicates the outlier factor; This represents the bias constant, assuming the LOF output is normalized and greater than or equal to 1. Set to 0; This represents the aging factor, which can be taken as 0.5; This indicates the current cumulative number of runs, with a value of 1 million. This indicates the design life, with a value of 5 million cycles.

[0067] If, under normal circumstances, , If the deduction is 0.1, then... point.

[0068] In abnormal circumstances, , If the denominator is 2.85, then point.

[0069] Finally, the comprehensive risk warning index is calculated. The comprehensive risk warning index satisfies the following relationship:

[0070]

[0071] In the formula, This represents the comprehensive risk warning index; This indicates the device's health score; Represents the conflict gain coefficient, for example Set to 20; This indicates the measured state of the micro switch (0 / 1). This indicates the theoretically expected state (0 / 1) inferred from the characteristics of the braking current sequence; 1 represents "action occurred / effective"; 0 represents "action stopped / ineffective".

[0072] In an optional embodiment, the method for determining the theoretically appropriate state includes:

[0073] Analyze the braking current sequence to identify whether there is a current inflection point that characterizes armature engagement;

[0074] If the current inflection point is detected, then the theoretically required state is determined to be the pull-in state;

[0075] If the current inflection point is not detected, then the theoretically required state is determined to be the release state.

[0076] The comprehensive risk warning index is compared with a preset alarm threshold; when the comprehensive risk warning index exceeds the preset alarm threshold, a shutdown protection operation is performed. For example, the preset alarm threshold is 100.

[0077] If it's just a decrease in health ( ), but the logic is consistent ( ),but The system prompts for maintenance, but does not force a shutdown.

[0078] If a logical conflict occurs, such as the current indicator being activated... But the switch did not move. At this point, health is extremely low (e.g., 40 points). At this point, the comprehensive risk warning index far exceeds the alarm threshold, triggering an immediate shutdown.

[0079] In this way, by combining quantitative health scores with logical conflict detection, we can achieve both visualized management of gradual failures and ensure absolute safety under severe failures.

[0080] like Figure 2 As shown, during the voltage dip period, the reduced voltage slows down the response, causing a sudden increase in the value above the alarm threshold, resulting in a false alarm. However, the normalized time curve, corrected by the formula of this invention, remains stable, proving that this invention effectively eliminates voltage interference.

[0081] like Figure 3 As shown, the normal current inflection point is deep and early, while the inflection point of mechanical hysteresis is gentle and later. The DTW algorithm of this invention can accurately capture this difference.

[0082] like Figure 4 As shown, in the two-dimensional feature space, normal data and corrected voltage fluctuation data are clustered in the safe area, while the real mechanical fault data are clearly isolated in the abnormal area, achieving zero false alarms and zero false alarms.

[0083] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0084] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A fault early warning method for gearless traction machines based on multi-source data fusion, characterized in that, include: Obtain the braking current sequence, driving voltage sequence, and micro switch state sequence during the braking process of the gearless traction mechanism, and extract the measured action time; The waveform distortion index between the braking current sequence and a preset standard fingerprint is calculated using a dynamic time warping algorithm to evaluate the morphological characteristics of the mechanical motion, including: During the elevator commissioning phase, the current waveforms of normal operation were recorded multiple times and the average waveform was generated as the preset standard fingerprint. Calculate the sum of the Euclidean distances between the currently acquired braking current sequence and the preset standard fingerprint after alignment, and use this sum as the waveform distortion index; A normalized action response time model is constructed. A voltage correction factor is calculated based on the driving voltage sequence, and a resistance correction factor is calculated based on the braking current sequence. The measured action time is corrected using the voltage correction factor and the resistance correction factor to obtain the normalized action time. The waveform distortion index and the normalized action time are input into the local outlier algorithm to obtain the outlier factor. The equipment health score is calculated based on the outlier factor. The comprehensive risk warning index is calculated by combining the logical consistency between the micro-switch state sequence and the braking current sequence, so as to realize the fault warning of the gearless traction machine.

2. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 1, characterized in that, Normalized action time satisfies the following relationship: In the formula, This represents the corrected normalized time; Indicates the actual measured action time; This is the measured average voltage; This is the rated voltage of the brake; This indicates the measured peak current. Indicates the reference current value; , This serves as a safety base to prevent the denominator from being zero. , The sensitivity index is set based on the characteristics of the electromagnetic coil.

3. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 1, characterized in that, The equipment health score satisfies the following relationship: In the formula, This indicates the device's health score; Indicates the outlier factor; Indicates the bias constant; Indicates the aging coefficient; This indicates the current cumulative number of runs; Indicates the design life.

4. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 1, characterized in that, The comprehensive risk warning index satisfies the following relationship: In the formula, This represents the comprehensive risk warning index; This indicates the device's health score; Indicates the conflict gain coefficient; Indicates the measured state of the micro switch; This indicates the theoretically expected state inferred from the characteristics of the braking current sequence.

5. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 2, characterized in that, The methods for obtaining the measured motion time include: Listen for the brake release command from the elevator main controller and record the time when the brake release command is issued as the start time; Monitor the state sequence of the microswitch and record the moment when the microswitch changes from closed to open as the flip-flop moment; The difference between the flipping time and the starting time is calculated to obtain the measured action time.

6. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 5, characterized in that, The measured average voltage is the arithmetic mean of the driving voltage sequence from the start time to the flip time; the peak current is the maximum value or steady-state holding current value reached by the braking current sequence within the acquisition window.

7. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 4, characterized in that, The methods for determining the theoretically required state include: Analyze the braking current sequence to identify whether there is a current inflection point that characterizes armature engagement; If the current inflection point is detected, then the theoretically required state is determined to be the pull-in state; If the current inflection point is not detected, then the theoretically required state is determined to be the release state.

8. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 1, characterized in that, The method further includes: The comprehensive risk warning index is compared with the preset alarm threshold; When the comprehensive risk warning index exceeds the preset alarm threshold, a shutdown protection operation is performed.

9. The fault early warning method for gearless traction machines based on multi-source data fusion according to claim 1, characterized in that, Set the sampling frequency to 5kHz; set the acquisition window duration to 400 milliseconds.

Images