A load monitoring method and system for a gear motor
By employing multi-time-delay scale coherence analysis and feature enhancement mechanisms, the problem of distinguishing between noise and load anomalies in existing technologies has been solved, enabling precise monitoring of the load status of gear motors and improving monitoring accuracy and anti-interference capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NINGBO JIANGBEI NEW XIN PETROCHENICAL MACHINERY EQUIP CO LTD
- Filing Date
- 2026-02-24
- Publication Date
- 2026-05-12
AI Technical Summary
Existing wavelet analysis methods based on current signals cannot effectively distinguish between noise and actual load anomalies in petrochemical valve gear motor load monitoring, resulting in insufficient monitoring accuracy and real-time performance. In particular, it is difficult to adaptively track the drift of load characteristic scales under strong noise and variable operating conditions.
A multi-delay scale coherence analysis mechanism is adopted. By obtaining the power spectral density and wavelet coefficients of the harmonic component sequence, the load fluctuation coherence index is calculated. Combined with feature enhancement gain and fusion weight, the feature recognition scale is adaptively adjusted to achieve accurate monitoring of load anomalies.
It significantly improves the accuracy and anti-interference capability of load monitoring, reduces the false alarm rate, and can output early warning and emergency shutdown commands in a timely manner to ensure the safe and stable operation of petrochemical production.
Smart Images
Figure CN121705974B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing. In particular, it relates to a method and system for monitoring the load of a gear motor. Background Technology
[0002] As a key actuator in petrochemical valve control systems, the load status of gear motors directly impacts production safety and system stability. In the petrochemical production environment, frequent valve starts, stops, and adjustments subject gear motors to complex and variable loads over extended periods, making them susceptible to load imbalances due to conditions such as unilateral jamming and uneven media flow. Failure to monitor and warn in a timely manner can lead to serious consequences, including valve control failure, media leakage, and even equipment damage, resulting in significant economic losses and safety risks. Therefore, accurate monitoring of gear motor load status is a necessary technical means to ensure the safe operation of petrochemical production.
[0003] Existing wavelet analysis methods based on current signals face a dual failure in petrochemical valve gear motor load monitoring under the coupled environment of strong noise and variable operating conditions: On the one hand, under strong electromagnetic interference, fixed threshold denoising strategies cannot distinguish between environmental noise such as power grid harmonics and random electromagnetic interference and weak harmonic characteristics caused by actual load anomalies, leading to early abnormal features being misjudged as noise and filtered out; on the other hand, manually preset fixed scale decomposition bands cannot adaptively track the load characteristic scale drift caused by variable operating conditions. Essentially, existing current wavelet analysis methods lack the ability to identify the randomness of noise and the regularity of load fluctuations, making it difficult to achieve adaptive tracking of characteristic scales, thus limiting the accuracy and real-time performance of load monitoring. Summary of the Invention
[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.
[0005] In a first aspect, the present invention provides a load monitoring method for a gear motor, comprising: acquiring a preset first number of time lags and a preset second number of scales; acquiring a harmonic component sequence in a preprocessed current sequence of the gear motor; taking any scale as a target scale; acquiring a power spectral density sequence of the target scale without time lag based on the harmonic component sequence; similarly acquiring a lag power spectral density sequence of the target scale at any time lag; calculating a load fluctuation coherence index of the target scale at any time lag based on the power spectral density sequence and the lag power spectral density sequence; and acquiring several optimal lags based on the load fluctuation coherence index. The process involves: taking the mean of all optimally lagging load fluctuation coherence indices as the comprehensive load fluctuation coherence at the target scale; calculating the feature enhancement gain at the target scale based on the comprehensive load fluctuation coherence; obtaining the wavelet coefficient sequence at the target scale and multiplying it by the feature enhancement gain at the target scale as the wavelet coefficient enhancement value sequence at the target scale; calculating the cosine similarity between the wavelet coefficient enhancement value sequences of any two scales in the scale set; calculating the fusion weight at the target scale based on the cosine similarity; calculating the final load anomaly score based on the fusion weight and cosine similarity; and completing load monitoring based on the final load anomaly score.
[0006] Preferably, obtaining the power spectral density sequence of the target scale without time lag based on the harmonic component sequence includes: performing wavelet transform based on the harmonic component sequence to obtain a wavelet coefficient sequence, calculating the square of the modulus of any wavelet coefficient to obtain the power spectral density, and smoothing all power spectral densities to obtain the power spectral density sequence of the target scale without time lag.
[0007] Preferably, the calculation of the load fluctuation coherence index at any time lag at the target scale includes: using the absolute value of the Pearson correlation coefficient between the power spectral density sequence and the lagged power spectral density sequence as the load fluctuation coherence index at any time lag at the target scale.
[0008] Preferably, obtaining several optimal lags based on load fluctuation coherence indices includes: traversing to obtain the load fluctuation coherence indices of the target scale power spectral density sequence without time lag and the target scale power spectral density sequence with lag at each time lag; for the target scale, sorting all load fluctuation coherence indices from largest to smallest, and selecting a preset number of time lags corresponding to load fluctuation coherence indices from largest to smallest as optimal lags.
[0009] Preferably, the calculation of the feature enhancement gain at the target scale includes: traversing to obtain the comprehensive load fluctuation coherence of each scale, and taking the scale corresponding to the maximum value of the comprehensive load fluctuation coherence as the center scale; calculating the first sum of the comprehensive load fluctuation coherence of the center scale and a preset constant, and calculating the first ratio of the comprehensive load fluctuation coherence of the target scale to the first sum; calculating the absolute difference between the target scale and the center scale, and calculating the second ratio of the absolute difference to the half-width, and calculating the negative exponent of the second ratio using an exponential function; calculating the first product of the negative exponent of the second ratio and the first ratio, and taking the sum of 1 and the first product as the feature enhancement gain at the target scale.
[0010] Preferably, the half-width includes: taking half of the overall load fluctuation coherence of the central scale as a condition parameter; taking the scale corresponding to the overall load fluctuation coherence not less than the condition parameter as a satisfying scale; calculating the difference between the maximum value and the minimum value of the scale in the satisfying scale, and taking half of the difference as the half-width.
[0011] Preferably, the calculation of the fusion weight for the target scale includes: traversing to obtain the cosine similarity between the target scale and the wavelet coefficient enhancement value sequence of any scale other than the target scale; using the sum of the cosine similarity between the target scale and the wavelet coefficient enhancement value sequence of each scale other than the target scale as a second sum; taking any two different scales as a combination, calculating the cosine similarity of any combination, and using the sum of the cosine similarity of all combinations as a third sum; and using the ratio of the second sum to the third sum as the fusion weight for the target scale.
[0012] Preferably, the calculation of the final load anomaly score includes: obtaining a wavelet coefficient enhancement value sequence at the target scale; taking the absolute value of each value in the wavelet coefficient enhancement value sequence to construct an absolute value sequence; calculating the maximum value and median value in the absolute value sequence; calculating the third ratio of the maximum value and median value; calculating the second product of the fusion weight at the target scale and the third ratio; traversing to obtain the second product at each scale; and accumulating the second products at all scales as the final load anomaly score.
[0013] Secondly, the present invention also provides a load monitoring system for a gear motor, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned load monitoring method for a gear motor is implemented.
[0014] The present invention has the following effects:
[0015] This invention effectively isolates noise and interference from current signals through wavelet transform and power spectral density analysis, focusing on the dynamic characteristics caused by load fluctuations, particularly periodic or transient disturbances caused by anomalies such as asymmetrical loads, mechanical jamming, and uneven flow. In noisy environments, an adaptive feature enhancement mechanism dynamically adjusts the feature recognition scale based on the comprehensive load fluctuation coherence, effectively addressing the time-frequency distribution drift of load characteristics and ensuring accurate capture of the actual load state. By fusing load fluctuation information captured at different scales, significant enhancement of load anomaly characteristics is achieved, significantly improving the sensitivity and anti-interference capability of early fault detection and reducing the false alarm rate. Finally, a multi-layered early warning mechanism based on comprehensive scoring can promptly output load fluctuation warnings, overload signals, or emergency shutdown commands, improving the accuracy and efficiency of load monitoring. Attached Figure Description
[0016] Figure 1 This is a flowchart of a load monitoring method for a gear motor according to an embodiment of the present invention. Detailed Implementation
[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.
[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] Reference Figure 1 A load monitoring method for a gear motor includes steps S1-S4, as detailed below:
[0020] S1: Obtain a preset first number of time lags and a preset second number of scales, obtain the harmonic component sequence in the preprocessed current sequence of the gear motor, take any scale as the target scale, obtain the power spectral density sequence of the target scale without time lag based on the harmonic component sequence, similarly obtain the lag power spectral density sequence of the target scale at any time lag, calculate the load fluctuation coherence index of the target scale at any time lag based on the power spectral density sequence and the lag power spectral density sequence, obtain several optimal lags based on the load fluctuation coherence index, and take the average of the load fluctuation coherence indices of all optimal lags as the comprehensive load fluctuation coherence of the target scale.
[0021] In one embodiment, a current sensor is installed at a critical location on the gear motor of the petrochemical valve control system to collect the current sequence during gear motor operation. The current sensor monitors the motor drive current non-invasively using the Hall effect principle. For example, the sampling frequency is set to 12.8 kHz according to the Nyquist sampling theorem to ensure signal integrity.
[0022] In the load monitoring of gear motors in petrochemical valves, the current signal comprises two parts: a fundamental frequency component and harmonic components. The fundamental frequency component mainly reflects the electromagnetic torque demand of the motor during steady-state operation, and its amplitude has an approximately linear relationship with the load size. However, it is not sensitive to asymmetric mechanical anomalies such as load imbalance and unilateral jamming. The harmonic components, on the other hand, originate from periodic or transient mechanical disturbances caused by gear meshing impacts, bearing microcracks, and uneven medium flow. These disturbances excite harmonic characteristics of specific frequencies in the current through electromagnetic-mechanical coupling effects, and their amplitude and phase changes directly characterize the symmetry and stability of the load distribution. Therefore, by separating the fundamental frequency from the current sequence through bandpass filtering and extracting the harmonic component sequence, steady-state operating information can be effectively removed, focusing on the weak dynamic characteristics caused by load anomalies. Further maximum-minimum value normalization processing makes the harmonic characteristics comparable within a unified numerical range.
[0023] It should be noted that in the high-noise environment of petrochemical plants, the core challenge of monitoring the load status of gear motors lies in the high degree of overlap between environmental noise and actual load fluctuations in the amplitude domain, making them difficult to distinguish directly. There are fundamental differences between the two: environmental noise (such as power grid harmonics and random electromagnetic interference) exhibits a wide frequency distribution and dispersed energy; its power spectral density fluctuates irregularly between adjacent time points and has extremely low correlation under multiple time delays, exhibiting typical random characteristics. In contrast, mechanical disturbances caused by load anomalies (such as valve jamming on one side or uneven medium flow) have clear scale-time locality; their energy is concentrated in a specific frequency band and exhibits stable high coherence over multiple time delays, reflecting the inherent regularity of load fluctuations. Traditional current amplitude analysis methods rely solely on amplitude thresholds for discrimination and cannot identify time-series fluctuation patterns, easily misjudging power grid interference as load anomalies. While fixed-scale wavelet analysis has time-frequency resolution capabilities, it relies on manually preset analysis frequency bands. When changes in operating conditions (such as changes in gear meshing characteristics caused by medium temperature fluctuations) lead to a systematic drift in the load characteristic scale, effective features are easily shifted out of the preset frequency band and missed. This invention proposes a multi-delay scale coherence analysis mechanism. By calculating the Pearson correlation coefficient of the wavelet power spectral density sequence of harmonic components over the entire time delay range, it adaptively searches for the optimal lag set that maximizes coherence. The comprehensive coherence index is used to quantify the intensity of the cross-delay law of load fluctuation, thereby achieving accurate identification of noise randomness and load regularity.
[0024] Obtain a preset first time lag and a preset second time scale. For example, the preset first time lag is 16 and the preset second time lag is 4. Specifically, the time lag is... , scale is .
[0025] Using any scale as the target scale, a continuous wavelet transform is performed on the preprocessed current harmonic component sequence to extract its time-frequency localization features at the target scale, obtaining a scale-time two-dimensional wavelet coefficient sequence. Then, the modulus square of the wavelet coefficients at each time point is calculated, and the complex domain coefficients are converted into real-valued energy density representations to form the original power spectral density sequence. This sequence intuitively reflects the instantaneous distribution of load fluctuation energy in the time-frequency domain. To suppress energy fluctuation spikes caused by measurement noise and random interference, a moving average filter is used to smooth the original power spectral density sequence, eliminating high-frequency random disturbances while retaining the regular energy changes caused by load anomalies, finally obtaining a smoothed power spectral density sequence without time lag at the target scale.
[0026] Similarly, the lagged power spectral density sequence at any time lag of the target scale is obtained. The absolute value of the Pearson correlation coefficient between the power spectral density sequence and the lagged power spectral density sequence is used as the load fluctuation coherence index at any time lag of the target scale.
[0027] By traversing a preset set of time lags, the Pearson correlation coefficient is calculated between the power spectral density sequence without lag at the same scale and the power spectral density sequence with any time lag, quantifying the self-similarity of load energy fluctuations at different time resolutions. When the load fluctuation coherence index is close to 1, it indicates that the energy at adjacent time points exhibits highly synchronized periodic changes, corresponding to regular mechanical disturbances caused by load imbalance (such as periodic shocks caused by valve jamming). When the load fluctuation coherence index is close to 0, it reflects that the energy fluctuations are completely random and disordered, corresponding to environmental noise or normal process fluctuations. The inherent normalization characteristic of the Pearson correlation coefficient makes it insensitive to the absolute amplitude of the signal, only responding to the intrinsic correlation of the fluctuation pattern, thus effectively avoiding amplitude interference such as power grid fluctuations and load size changes; while the absolute value processing further strictly limits the value to between 0 and 1, only considering whether there is a correlation.
[0028] For any target scale, the load fluctuation coherence indexes of the power spectral density sequence without time lag and the power spectral density sequence with time lag at each time lag of the target scale are obtained through iteration, forming a complete time delay-coherence mapping relationship at that scale; then, all load fluctuation coherence indices are arranged in descending order, and a preset number of time lags corresponding to the load fluctuation coherence indices are selected as the optimal lags from largest to smallest. For example, four are selected from largest to smallest, and the specific number can be set by those skilled in the art.
[0029] It should be explained that real load anomalies (such as valve jamming or uneven media) will exhibit high coherence synchronously under multiple time delays due to their inherent periodicity or quasi-periodicity, while the high coherence of environmental noise only occasionally appears in individual time delays. By filtering and aggregating information from multiple high-coherence time delays, the stability of load imbalance characteristics can be effectively enhanced, while suppressing random fluctuations caused by instantaneous power grid interference or measurement noise under a single time delay.
[0030] The mean of all optimal lag load fluctuation coherence indices is used as the target scale for comprehensive load fluctuation coherence.
[0031] S2: Calculate the feature enhancement gain at the target scale based on the coherence of the comprehensive load fluctuation.
[0032] It should be noted that in the petrochemical production process, frequent adjustments in the valve control system cause dynamic changes in the load conditions of the gear motors, resulting in a systematic scale drift in the time-frequency distribution of abnormal load characteristics. On the one hand, fluctuations in medium temperature cause thermal expansion or contraction of the gearbox, altering the gear meshing stiffness and transmission clearance, causing the instantaneous jamming impact characteristics, originally concentrated at the high-frequency scale, to migrate to the mid-frequency scale. On the other hand, changes in load magnitude alter the gear meshing load distribution, causing the energy center of gravity of periodic fluctuation characteristics to shift within the scale domain. Traditional fixed-parameter feature enhancement methods rely on manually preset static scale bands and cannot detect the above drift. When the actual abnormal characteristics move out of the preset frequency band, they are weakened, while the noise-dominant frequency band is incorrectly enhanced, resulting in false alarms or missed detections. This invention, through condition-adaptive feature enhancement gain, automatically identifies the feature scale position and dynamically adjusts the enhancement area, achieving precise focusing of load imbalance characteristics.
[0033] In one embodiment, the overall load fluctuation coherence of each scale is obtained by traversal, and the scale corresponding to the maximum value of the overall load fluctuation coherence is taken as the central scale.
[0034] Due to the differences in their temporal evolution characteristics, different types of load anomalies (such as instantaneous shocks, periodic fluctuations, and slow changes) will exhibit the most significant energy concentration and temporal regularity at a specific scale. When the analysis scale matches the inherent time scale of the anomaly characteristics, its coherence index will naturally reach its peak. Obtaining the central scale can adaptively identify the scale at which the most significant load characteristics under the current operating conditions are located.
[0035] The algorithm calculates the first sum of the comprehensive load fluctuation coherence at the center scale and a preset constant, and calculates the first ratio of the comprehensive load fluctuation coherence at the target scale to the first sum. It then calculates the absolute difference between the target scale and the center scale, and calculates the second ratio of the absolute difference to the half-width, using an exponential function to calculate the negative exponent of the second ratio. Finally, it calculates the first product of the negative exponent of the second ratio and the first ratio, and uses the sum of 1 and the first product as the feature enhancement gain at the target scale. The preset constant is used to prevent the denominator from being zero; an exemplary value is 0.00001.
[0036] The half-width includes: using half of the overall load fluctuation coherence of the central scale as a condition parameter; using the scale corresponding to an overall load fluctuation coherence not less than the condition parameter as a satisfying scale; calculating the difference between the maximum and minimum values of the satisfying scales, and using half of the difference as the half-width. For example, if the central scale is 8, the overall load fluctuation coherence of scale 4 is 0.3, that of scale 8 is 0.9, that of scale 16 is 0.6, and that of scale 32 is 0.2, and the condition parameter is half of 0.9, i.e., 0.45, then scales 8 and 16 are satisfying scales, and the half-width is half the difference between 16 and 8, i.e., 4.
[0037] For the feature enhancement gain, 1 represents the baseline gain, ensuring that the wavelet coefficients at all analysis scales maintain at least their original amplitude, avoiding distortion of the effective signal due to excessive suppression. Based on this, the gain is dynamically adjusted according to the first ratio of each scale to the center scale: when the overall load fluctuation coherence at any scale is close to that at the center scale, the first ratio approaches one, and the feature enhancement gain increases to nearly double, significantly amplifying weak features caused by load imbalance, such as periodic impacts from valve jamming or regular fluctuations caused by media inhomogeneity; conversely, when the coherence at any scale is far lower than that at the center scale, the first ratio approaches zero, and the feature enhancement gain falls back to the baseline level, effectively suppressing useless components in the noise-dominated frequency bands such as grid harmonics and electromagnetic interference. Simultaneously, when changes in operating conditions cause load feature scale drift, the center scale position is dynamically updated accordingly, and the entire enhancement region automatically migrates and refocuses on the frequency band containing the most significant abnormal features under the current operating conditions, achieving adaptive feature enhancement based on operating conditions.
[0038] S3: Obtain the wavelet coefficient sequence at the target scale, and use the product of the wavelet coefficient sequence and the feature enhancement gain at the target scale as the wavelet coefficient enhancement value sequence at the target scale.
[0039] In one embodiment, the wavelet coefficient sequence at the target scale can be obtained according to step S1, and the product of the wavelet coefficient sequence and the feature enhancement gain at the target scale is used as the wavelet coefficient enhancement value sequence at the target scale.
[0040] S4: Calculate the cosine similarity of wavelet coefficient enhancement value sequences of any two scales in the scale set, calculate the fusion weight of the target scale based on the cosine similarity, calculate the final load anomaly score based on the fusion weight and cosine similarity, and complete the load monitoring based on the final load anomaly score.
[0041] It should be noted that wavelet transform can capture complementary features of gear motor load fluctuations at different scales: high-frequency scales are sensitive to instantaneous impacts, while mid- and low-frequency scales characterize periodic or slow-changing processes. However, traditional fusion methods, which employ fixed weights or simple averaging strategies, cannot distinguish between real load imbalance features and environmental noise. In the context of strong interference in petrochemical environments, grid harmonics and electromagnetic interference exhibit random distribution, low coherence, and uncorrelated feature patterns across scales at all scales. In contrast, the abnormal features caused by real load imbalances (such as valve jamming or uneven medium flow) not only exhibit high coherence at specific scales but also maintain highly consistent temporal fluctuation patterns across adjacent scales. This invention proposes an adaptive fusion mechanism based on scale consistency. By automatically assigning greater fusion weights to highly consistent scales, the final anomaly score focuses on reliable features jointly verified by multiple scales, while suppressing random interference from isolated scales. This significantly improves the sensitivity and false alarm resistance of early load imbalance monitoring in strong noise environments.
[0042] In one embodiment, the cosine similarity between wavelet coefficient enhancement value sequences corresponding to any two different scales is calculated to quantify the consistency of load fluctuation features captured at each scale in terms of temporal morphology. For the target scale, the cosine similarity between the target scale and any other scale is summed to form a second sum representing the overall synergy between the target scale and other scales. Simultaneously, the cosine similarities corresponding to pairwise combinations of all scales are globally summed to form a third sum reflecting the consistency of features across all scales. The ratio of the second sum to the third sum is used as the fusion weight for the target scale.
[0043] The logic of weight fusion is as follows: when the target scale shows high similarity with most other scales (indicating that the target scale captures genuine load anomaly features jointly verified by other scales), the weight automatically increases; when the target scale features differ significantly from other scales (indicating that the target scale may be dominated by noise), the weight is naturally suppressed. Simultaneously, the normalization mechanism ensures that the sum of the weights of all scales remains constant, so that the final anomaly score focuses on reliable features jointly verified by multiple scales while avoiding interference from single-scale anomaly fluctuations on the overall judgment.
[0044] A wavelet coefficient enhancement value sequence at the target scale is obtained. The absolute value of each value in the wavelet coefficient enhancement value sequence is taken to construct an absolute value sequence. The maximum and median values in the absolute value sequence are calculated, and the third ratio of the maximum and median values is calculated to characterize the significance of the anomalous features at that scale: Under normal operating conditions, the wavelet coefficients are uniformly distributed, the maximum and median values are close, and the ratio approaches one; when there is load imbalance, abnormal shocks or periodic fluctuations will produce significant peaks in the time domain, making the maximum value significantly larger than the median value, and the ratio increases accordingly. The third ratio is further multiplied by the fusion weight of the target scale, where the fusion weight is obtained by normalizing the cosine similarity between scales. This automatically amplifies reliable anomalous signals that are highly consistent with the feature morphology of other scales, while suppressing random noise interference from isolated scales. The second product of each scale is accumulated across all scales to form the final load anomaly score. This score comprehensively reflects the intensity of the anomalous features at multiple scales and the cross-scale consistency; a higher value indicates a more severe load imbalance.
[0045] Load monitoring is performed based on the final load anomaly score. For example, if the final load anomaly score is no greater than 1.2, no alarm is triggered, indicating that the load is stable and there are no significant abnormal characteristics. If the final load anomaly score is greater than 1.2 but not greater than 1.8, a load fluctuation warning signal is output, indicating that there is early wear in gear meshing or bearings, causing periodic load fluctuations. If the final load anomaly score is greater than 1.8 but not greater than 2.5, an overload warning signal is output, indicating that the load anomaly has affected the system stability and that valve openings need to be adjusted or operating speed reduced. If the final load anomaly score is greater than 2.5, an emergency stop signal is output, indicating that the load is severely unbalanced and there is a risk of mechanical damage.
[0046] The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement a load monitoring method for a gear motor according to the first aspect of the present invention.
[0047] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.
[0048] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.
Claims
1. A method for monitoring the load of a geared motor, characterized in that, include: The process involves obtaining a preset first number of time lags and a preset second number of scales, obtaining the harmonic component sequence in the preprocessed current sequence of the gear motor, taking any scale as the target scale, and obtaining the power spectral density sequence of the target scale without time lag based on the harmonic component sequence. This includes: performing wavelet transform based on the harmonic component sequence to obtain the wavelet coefficient sequence, calculating the square of the modulus of any wavelet coefficient to obtain the power spectral density, and smoothing all power spectral densities to obtain the power spectral density sequence of the target scale without time lag. Similarly, based on the calculation method of the power spectral density sequence, the lagged power spectral density sequence of the target scale at any time lag is obtained. The load fluctuation coherence index of the target scale at any time lag is calculated based on the power spectral density sequence and the lagged power spectral density sequence. Several optimal lags are obtained based on the load fluctuation coherence index. The average of the load fluctuation coherence indices of all optimal lags is taken as the comprehensive load fluctuation coherence of the target scale. The feature enhancement gain at the target scale is calculated based on the overall load fluctuation coherence. Obtain the wavelet coefficient sequence at the target scale, and multiply the wavelet coefficient sequence with the feature enhancement gain at the target scale as the wavelet coefficient enhancement value sequence at the target scale. The cosine similarity of wavelet coefficient enhancement value sequences of any two scales in the scale set is calculated. The fusion weight of the target scale is calculated based on the cosine similarity. The final load anomaly score is calculated based on the fusion weight and cosine similarity. Load monitoring is completed based on the final load anomaly score. The feature enhancement gain calculated at the target scale includes: The overall load fluctuation coherence of each scale is obtained by traversing the scale, and the scale corresponding to the maximum value of the overall load fluctuation coherence is taken as the central scale. Calculate the first sum of the comprehensive load fluctuation coherence at the central scale and a preset constant, and calculate the first ratio of the comprehensive load fluctuation coherence at the target scale to the first sum. Calculate the absolute difference between the target scale and the center scale, and calculate the second ratio of the absolute difference to the half-width. Use an exponential function to calculate the negative exponent of the second ratio. Calculate the negative exponent of the second ratio and the first product of the first ratio, and use the sum of 1 and the first product as the feature enhancement gain for the target scale.
2. The load monitoring method for a gear motor according to claim 1, characterized in that, The load fluctuation coherence indexes for the target scale at any time lag include: The absolute value of the Pearson correlation coefficient between the power spectral density sequence and the lagged power spectral density sequence is used as the load fluctuation coherence index at any time lag for the target scale.
3. The load monitoring method for a gear motor according to claim 1, characterized in that, The method of obtaining several optimal hysteresis based on load fluctuation coherence index includes: The load fluctuation coherence index is obtained by iterating through the power spectral density sequence of the target scale without time lag and the power spectral density sequence of the target scale with time lag at each time lag. For the target scale, all load fluctuation coherence indices are sorted from largest to smallest, and a preset number of load fluctuation coherence indices corresponding to the time lags are selected as the optimal lags.
4. The load monitoring method for a gear motor according to claim 1, characterized in that, The half-width includes: Half of the overall load fluctuation coherence at the central scale is used as the conditional parameter; The scale corresponding to the comprehensive load fluctuation coherence that is not less than the condition parameter is used as the satisfaction scale. Calculate the difference between the maximum and minimum scale values in the scale, and take half of the difference as the half-width.
5. The load monitoring method for a gear motor according to claim 1, characterized in that, The fusion weights for calculating the target scale include: The cosine similarity between the target scale and the wavelet coefficient enhancement value sequence of any scale other than the target scale is obtained by iterating through the data. The sum of the cosine similarity between the target scale and the wavelet coefficient enhancement value sequence of each scale other than the target scale is used as the second sum. Take any two different scales as a combination, calculate the cosine similarity of any combination, and take the sum of the cosine similarities of all combinations as the third sum. The ratio of the second sum to the third sum is used as the fusion weight for the target scale.
6. The load monitoring method for a gear motor according to claim 1, characterized in that, The calculation of the final load anomaly score includes: Obtain the wavelet coefficient enhancement value sequence at the target scale, take the absolute value of each value in the wavelet coefficient enhancement value sequence to construct the absolute value sequence, calculate the maximum and median values in the absolute value sequence, and calculate the third ratio of the maximum and median values. Calculate the second product of the fusion weights at the target scale and the third ratio; Iterate through each scale to obtain the second product, and sum the second products of all scales as the final load anomaly score.
7. A load monitoring system for a geared motor, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a load monitoring method for a gear motor according to any one of claims 1-6.