An automated stereoscopic warehouse equipment fault diagnosis method and system

Abstract

The application belongs to the technical field of equipment monitoring, and particularly relates to an automatic stereoscopic warehouse equipment fault diagnosis method and system, which comprises the following steps: collecting time sequence data segments of the automatic stereoscopic warehouse equipment; extracting a mixed feature vector containing regular statistical features and time sequence shape features from the data segments; preliminarily judging the fault nature of the time sequence data segments based on the time sequence shape features to obtain a shape category; determining the diagnosis weight of the regular statistical features according to the shape category, and calculating the adaptive weighted distance between the regular statistical features of the current data segment and a preset fault prototype to determine the final fault diagnosis result. By introducing time sequence shape analysis and dynamic feature weighting, the application can effectively distinguish fault modes that are similar in statistics but different in physical causes, and significantly improves the accuracy of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of equipment monitoring technology. More specifically, this invention relates to a method and system for diagnosing faults in automated storage and retrieval systems (AS / RS) equipment. Background Technology

[0002] In modern intelligent warehousing and logistics systems, stacker cranes are key equipment for accessing goods on high-level racks, and their highly reliable operation is fundamental to ensuring the efficiency and effectiveness of the entire logistics center. As a complex electromechanical system integrating motors, gearboxes, transmission mechanisms, and control units, any early failure in any component of the stacker crane lifting system, if not detected in time, can escalate into a major malfunction, causing significant economic losses and safety risks.

[0003] Currently, the mainstream technical approach involves installing sensors, such as motor current transformers, vibration accelerometers, and position encoders, on key components of the stacker crane to synchronously collect equipment operation data. Then, within preset timescale windows, a series of statistical and frequency domain features are extracted from the raw time-series signals of each sensor. These multi-scale features are then simply concatenated into a high-dimensional feature vector. Using this high-dimensional feature vector and corresponding historical fault labels, a machine learning classification model, such as a decision tree or random forest, is trained to identify and attribute faults in real-time data.

[0004] However, using statistical aggregation to characterize signals within a time window inevitably compresses a dynamic signal with specific waveforms and evolutionary trends into a set of static values, completely losing the signal's temporal morphology information within the window. For example, within the same time window, two signals with vastly different morphologies—a slow, linear rise in current caused by poor lubrication of the transmission chain and a sudden, pulse-like spike in current caused by transient interference in the electrical system—may have extremely similar mean, maximum, and energy values. Since decision trees and other models rely on these highly abstracted static statistical values, they cannot effectively distinguish between these two fault precursors, which are determined by the signal process and morphology and represent different physical causes. This severely limits the precision and accuracy of fault diagnosis, especially in the identification of early-stage faults, easily leading to missed or misjudged cases. Summary of the Invention

[0005] To address the technical problem of limited diagnostic accuracy caused by the loss of signal timing and morphological information in existing technologies, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a fault diagnosis method for automated storage and retrieval system (AS / RS) equipment, comprising: acquiring multi-source sensor data from the AS / RS equipment and dividing the acquired continuous time-series data stream into multiple time-series data segments; extracting a hybrid feature vector for each time-series data segment, wherein the hybrid feature vector includes conventional statistical features describing the overall energy and distribution of the signal, and time-series morphological features characterizing the waveform shape of the signal within the time-series data segment; performing a preliminary judgment on the fault nature of the time-series data segment based on the time-series morphological features to obtain the morphological category corresponding to the time-series data segment; determining the diagnostic weight of the conventional statistical features according to the morphological category; calculating the adaptive weighted distance between the conventional statistical features of the current time-series data segment and a preset set of multiple fault prototypes; and determining the final fault diagnosis result based on the fault prototype with the smallest adaptive weighted distance.

[0007] This invention, by introducing time-series morphological analysis and dynamic feature weighting, can effectively distinguish fault modes that are similar in traditional statistics but have different physical causes, thus significantly improving the precision and accuracy of fault diagnosis.

[0008] Preferably, the time series morphological features include: an instantaneous impact index, used to assess the degree of pulse spikes in the time series data segment; a morphological skew index, used to assess the symmetry of the energy distribution of the time series data segment over time; and a trend stability index, used to assess the degree to which the time series data segment exhibits a monotonic trend or high-frequency oscillations.

[0009] By introducing these three time-series morphological characteristics, the dynamic morphology of the signal can be fully captured from three dimensions: impact, time symmetry, and trend, providing a rich and physically meaningful information basis for subsequent fault nature determination.

[0010] Preferably, the instantaneous impact index satisfies the following relationship: ;in, For instantaneous impact index, For the first time series data segment The value of each sampling point, This represents the total number of sampling points within the time-series data segment.

[0011] Preferably, the morphological skewness index satisfies the following relationship: ;in, This is the morphological skewness index. For the first time series data segment The value of each sampling point, This represents the total number of sampling points within the time-series data segment.

[0012] By calculating the morphological skew index, it is possible to effectively determine whether abnormal fluctuations occur in the first or second half of the window. This is crucial for distinguishing between shocks during the startup phase and jitters during the shutdown phase, which are related to specific operational sequences, thus improving the precision of diagnosis.

[0013] Preferably, the trend stability index satisfies the following relationship: ;in, As a trend stability index, , For symbolic functions, For the first time series data segment The value of each sampling point, This represents the total number of sampling points within a time-series data segment. This represents the total number of local trend changes within the entire time series data segment.

[0014] By calculating the trend stability index, it is possible to effectively assess whether the signal exhibits a monotonic trend or high-frequency oscillations, thereby distinguishing different fault types such as continuous current rise caused by abnormal load and high-frequency oscillations caused by mechanical resonance.

[0015] Preferably, the step of making a preliminary judgment on the fault nature of the time-series data segment based on the time-series morphological features to obtain the morphological category corresponding to the time-series data segment includes: combining the time-series morphological features extracted from all sensor signals into a multi-dimensional morphological feature vector; classifying the multi-dimensional morphological feature vector using a pre-trained rule discriminator to obtain the morphological category corresponding to the time-series data segment, wherein the morphological category includes impact type, oscillation type and trend type.

[0016] Preferably, the specific process of determining the diagnostic weight of the conventional statistical feature based on the morphological category is as follows: a corresponding diagnostic feature weight vector is pre-constructed for each morphological category, and the diagnostic weight of the conventional statistical feature is determined based on the diagnostic feature weight vector.

[0017] By using dynamic weighting, the diagnostic model can adaptively focus on the most relevant feature dimensions based on the macroscopic nature of the fault, thus achieving a focus on specific faults and greatly improving the accuracy of fault attribution.

[0018] Preferably, the adaptive weighted distance satisfies the following relationship: ;in, This is the regular statistical feature vector of the current time series data segment. This is the fault prototype vector. The dimension of a conventional statistical feature vector. Based on morphological categories The determined diagnostic feature weight vector of the first The weights of each feature, and These are the first two elements in the real-time feature vector and the fault prototype vector, respectively. Each feature value.

[0019] Preferably, the conventional statistical features include at least one of the following: mean, root mean square value, variance, skewness, kurtosis, and kurtosis.

[0020] Secondly, the present invention provides an automated storage and retrieval system fault diagnosis system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned automated storage and retrieval system fault diagnosis method is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned method for diagnosing faults in automated storage and retrieval systems, and stored in a memory for loading and execution by a processor. This allows for the creation of terminal equipment based on the memory and processor, making it convenient to use.

[0022] This invention assesses the morphological characteristics of signals within a time window, such as their impulsiveness, skewness, and trend. This enables diagnostic models to effectively distinguish between fault modes that are similar in traditional statistics but have completely different physical causes, greatly improving the precision and accuracy of fault attribution. The morphological characteristic index proposed in this invention is more sensitive to subtle changes, thus enabling earlier detection of the nascent stage of a fault and providing a valuable time window for predictive maintenance.

[0023] Furthermore, by using the preliminary discrimination results of signal morphology to dynamically guide the subsequent feature weight allocation, the model can adaptively focus on the most relevant statistical feature dimensions when facing different types of anomalies. Attached Figure Description

[0024] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein:

[0025] Figure 1 This is a flowchart illustrating a fault diagnosis method for automated storage and retrieval system equipment according to the present invention;

[0026] Figure 2 This is a schematic diagram illustrating the signal waveforms under four different operating conditions;

[0027] Figure 3 This is a schematic diagram illustrating the distribution of data points for four operating conditions in a three-dimensional feature space. Detailed Implementation

[0028] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] This invention discloses a fault diagnosis method for automated storage and retrieval system (AS / RS) equipment, referring to... Figure 1 This includes steps S1-S4:

[0031] S1. Collect multi-source sensor data from the automated storage and retrieval system (AS / RS) equipment and divide the collected continuous time-series data stream into multiple time-series data segments.

[0032] In one optional embodiment, various types of sensors are deployed at key monitoring locations of the stacker crane. Specifically, high-frequency triaxial accelerometers are deployed at key locations such as the housing of the lifting motor, the gearbox housing, and the column guide rails to monitor the vibration status of the equipment; Hall current sensors are deployed in the main power supply circuit of the lifting motor to monitor the load and electrical health of the motor; and the high-precision position encoder built into the servo system is used to obtain the precise position information of the lifting platform.

[0033] In this optional embodiment, a preset sampling frequency, exemplarily set to 10kHz, is used to synchronously acquire multi-source heterogeneous data of the stacker crane throughout its complete operational cycle, including stages such as start-up, rising, falling, and stopping. To ensure the synchronization of data from each sensor, all acquisition channels use a unified hardware clock or are timestamped via Network Time Protocol (NTP) to ensure strict consistency of data in time.

[0034] Furthermore, the acquired continuous time-series data stream is sliced ​​according to a fixed time window size W (exemplarily set to 100ms) and a certain overlap rate (exemplarily set to 50%). The 100ms window size ensures that most transient impact events are captured while providing sufficient data points for subsequent frequency domain analyses such as Fourier transforms, thus guaranteeing the accuracy of the analysis. Through this processing, the continuous raw data stream is converted into a series of multi-channel time-series data segments.

[0035] Thus, high-frequency synchronous acquisition and reasonable window division provide a high-quality data foundation for subsequent feature extraction and diagnostic analysis.

[0036] S2. Extract a hybrid feature vector for each time-series data segment. The hybrid feature vector includes conventional statistical features that describe the overall energy and distribution of the signal, as well as time-series morphological features that characterize the waveform shape of the signal within the time-series data segment.

[0037] In an optional embodiment, for each sensor signal, its conventional statistical characteristics within a time window are calculated to form a statistical feature vector. These characteristics are macroscopic indicators describing the overall energy and distribution of the signal, specifically including but not limited to:

[0038] (1) Mean: Reflects the DC component or baseline level of the signal.

[0039] (2) Root mean square (RMS): Reflects the effective value or average power of the signal.

[0040] (3) Variance: assesses the degree of fluctuation of the signal around its mean.

[0041] (4) Skewness: assesses the asymmetry of the signal probability distribution.

[0042] (5) Kurtosis: assesses the thickness of the peaks or tails of the signal probability distribution, and is sensitive to impulsive signals.

[0043] (6) Peak-to-peak value: The difference between the maximum and minimum values ​​of a signal, reflecting the dynamic range of the signal.

[0044] In this optional embodiment, a temporal morphology feature characterizing the waveform shape of a signal within a time-series data segment can be constructed, including three temporal morphology feature indices, which are used to evaluate the waveform characteristics of the signal within the window from different dimensions, forming a morphology feature vector.

[0045] Specifically, a stable signal and a signal containing transient impulses, even if their root mean square values ​​are similar, have completely different physical meanings. Transient impulses often correspond to sudden faults such as electrical arcing or mechanical impacts. To assess the intensity of the signal's pulse spikes, the transient impulse index can be calculated, which satisfies the following relationship:

[0046]

[0047] in, For instantaneous impact index, For the first time series data segment The value of each sampling point, This represents the total number of sampling points within a time-series data segment. This represents the maximum absolute value of the signal within the window, which is the peak value of the signal. The denominator is the root mean square value of the signal, and the exponent is the ratio of the peak value to the effective value of the signal.

[0048] In this optional embodiment, for a periodic signal with a stable shape, such as a sine wave, there is a fixed proportional relationship between the peak value and the effective value. However, when there is a very sharp pulse in the signal that is much higher than the average level, the peak value will be much larger than the overall effective value of the signal, resulting in... The value is very large, therefore The larger the value, the stronger the impulse characteristics of the signal.

[0049] In an optional embodiment, to assess the symmetry of signal energy distribution within a time window—that is, to determine whether abnormal fluctuations occur in the first half, second half, or are uniformly distributed within the window—and thus distinguish between shocks during the startup phase and jitters during the shutdown phase, which are faults related to specific operational timing, a morphological skew index can be calculated. The morphological skew index satisfies the following relationship:

[0050]

[0051] in, The morphological skewness index calculates the normalized difference between the energy in the first half and the second half of the window, and its range is [value range missing]. When the majority of the signal energy is concentrated in the first half of the window, the exponent value tends to +1; when the energy is concentrated in the second half, the exponent value tends to -1; when the energy is evenly distributed in both halves, the exponent value also tends to 0.

[0052] In an optional embodiment, to evaluate whether the signal exhibits a monotonic trend (continuous rise or fall) or high-frequency oscillation within a window, for example, an abnormal load may cause the current signal to rise slowly and continuously, while mechanical resonance manifests as high-frequency oscillation, a trend stability index can be calculated. The trend stability index satisfies the following relationship:

[0053]

[0054] in, As a trend stability index, , It is a symbolic function, and its value is... These represent decline, stability, and rise, respectively. This represents the total number of local trend changes within the entire time series data segment.

[0055] In this optional embodiment, the value range of the trend stability index is: An ideal monotonic signal has an unchanging trend, a summation term of 0, and an exponent of 1; an extreme noisy signal that changes direction at every sampling point has a maximum summation term and an exponent that tends towards 0. Therefore, The larger the value, the stronger the trend of the signal.

[0056] Finally, the obtained statistical feature vectors and morphological feature vectors are combined to form a hybrid feature vector.

[0057] Thus, by combining conventional statistical features with temporal morphological features, a hybrid feature vector that can comprehensively describe the macroscopic statistical characteristics and microscopic dynamic morphology of a signal is constructed, laying a solid foundation for subsequent accurate diagnosis.

[0058] S3. Based on the temporal morphological characteristics, the fault nature of the time-series data segment is initially determined, and the morphological category corresponding to the time-series data segment is obtained.

[0059] In an optional embodiment, a pre-trained rule discriminator can be used to classify multidimensional morphological feature vectors to obtain the morphological category of abnormal signals within the current time window, wherein the morphological category includes impact type, oscillation type and trend type.

[0060] Specifically, the discrimination rules included in the rule discriminator are: Rule 1: If the instantaneous current impact index Or instantaneous impact index of vibration The morphological type is impact type, indicating that there is a significant sudden spike in the signal. Rule 2: If the vibration trend stability index If the morphology is oscillating, it indicates that the vibration signal exhibits high-frequency fluctuations without a clear trend. Rule 3: If the impact index is low and the current trend stability index is low... If the current signal is stable and exhibits a clear upward or downward trend, then the pattern category is trend-based, indicating a stable current signal with a clear upward or downward trend. Rule 4: If none of the above conditions are met, then the pattern category is normal.

[0061] For example, suppose the morphological feature vector calculated from the current and vibration signals within a time window is: ,because According to rule 1, the system initially classifies the window's shape as impact type.

[0062] In this way, by making a preliminary judgment on the signal pattern, the complex fault diagnosis problem is broken down into simpler and more targeted sub-problems, providing important prior information for the next stage of fine attribution.

[0063] S4. Determine the diagnostic weight of conventional statistical features based on the morphological category, calculate the adaptive weighted distance between the conventional statistical features of the current time series data segment and the preset multiple fault prototypes, and determine the final fault diagnosis result based on the fault prototype with the smallest adaptive weighted distance.

[0064] In an optional embodiment, a fault knowledge base can be constructed. During the training phase, for each known specific fault type, such as motor overload, bearing outer ring damage, and normal operating conditions, a large amount of data is collected and conventional statistical feature vectors are extracted. Their cluster centers are calculated to form fault prototype vectors.

[0065] Furthermore, a corresponding diagnostic feature weight vector is pre-constructed for each morphological category. The dimension of this weight vector is the same as that of the statistical feature vector. The core objective of weight setting is to highlight the statistical features most relevant to the morphology.

[0066] Specifically, if the pattern type is impact-type, the weight vector will amplify the weights of features such as peak-to-peak value and kurtosis, because these features best reflect the intensity of the impact. For example, the weights can be set to {mean: 0.1, RMS: 0.1, variance: 0.2, kurtosis: 0.9, peak-to-peak value: 0.9}. If the pattern type is trend-type, the weight vector will amplify the weights of features such as mean and root mean square value to determine the degree of deviation of the signal baseline. For example, the weights can be set to {mean: 0.9, RMS: 0.8, variance: 0.2, kurtosis: 0.1, peak-to-peak value: 0.3}. If the pattern type is oscillation-type, the weight vector will amplify the weights of variance and statistical features related to the oscillation signal.

[0067] In an optional embodiment, after determining the morphology category, the system immediately calls the corresponding diagnostic feature weight vector, and then calculates the adaptive weighted Euclidean distance between the statistical feature vector of the current window and all fault prototype vectors in the knowledge base. The adaptive weighted distance satisfies the following relationship:

[0068]

[0069] in, This is the regular statistical feature vector of the current time series data segment. This is the fault prototype vector. The dimension of a conventional statistical feature vector. Based on morphological categories The determined diagnostic feature weight vector of the first The weights of each feature, and These are the first two elements in the real-time feature vector and the fault prototype vector, respectively. Each feature value.

[0070] Furthermore, the fault prototype vector with the smallest weighted distance is calculated, and its corresponding fault label is the final fine-grained attribution result, thereby determining the final fault diagnosis result.

[0071] In an optional embodiment, such as Figure 2The diagram shown is a schematic representation of signal waveforms under four different operating conditions in an embodiment of the present invention, including normal, overload, jam, and wear. As can be seen, Figure 2 Under medium overload conditions, the signal shows a slow upward trend; under jamming conditions, there are violent instantaneous spikes; and under wear conditions, there are obvious high-frequency oscillation components superimposed. These differences intuitively prove that different faults have unique timing patterns.

[0072] In an optional embodiment, such as Figure 3 The diagram illustrates the distribution of data points for four operating conditions in a three-dimensional feature space, as shown in this embodiment of the invention. The feature space comprises the instantaneous current impact index, the vibration trend stability index, and the mean current. As can be seen, the data points for the four states form four distinct clusters: the stuck state, due to its massive current impact, is distributed in the high-value region of the X-axis; the wear state, due to its severe vibration, is uniquely distributed in the low-value region of the Y-axis; the overload state, due to its significant increase in the mean current, is distributed in the high-value region of the Z-axis; and the normal state data is clustered in the region near the origin. This diagram visually verifies that the multi-dimensional space constructed by this invention, which integrates morphological and statistical features, can effectively separate various operating conditions with drastically different physical meanings.

[0073] Thus, by using dynamic weighting and distance discrimination based on morphological categories, faults with similar statistical characteristics but different morphological characteristics can be accurately distinguished, significantly improving the accuracy and reliability of diagnosis.

[0074] This invention also discloses an automated storage and retrieval system (AS / RS) equipment fault diagnosis system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an automated storage and retrieval system fault diagnosis method according to the present invention is implemented.

[0075] 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.

[0076] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise expressly and specifically defined.

[0077] While this specification has shown and described numerous embodiments of the invention, 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. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A fault diagnosis method for automated storage and retrieval system equipment, characterized in that, include: Collect multi-source sensor data from automated storage and retrieval systems (AS / RS) equipment and divide the collected continuous time-series data stream into multiple time-series data segments; For each time-series data segment, a hybrid feature vector is extracted. The hybrid feature vector includes conventional statistical features that describe the overall energy and distribution of the signal, as well as time-series morphological features that characterize the waveform shape of the signal within the time-series data segment. The temporal morphological characteristics include: the instantaneous impact index, which is used to assess the degree of pulse spikes in a time series data segment; the morphological skew index, which is used to assess the symmetry of the energy distribution of a time series data segment over time; and the trend stability index, which is used to assess the degree to which a time series data segment exhibits a monotonic trend or high-frequency oscillations. Based on the temporal morphological features, the fault nature of the time series data segment is initially determined, and the morphological category corresponding to the time series data segment is obtained. This includes: combining the temporal morphological features extracted from all sensor signals into a multi-dimensional morphological feature vector; and classifying the multi-dimensional morphological feature vector using a pre-trained rule discriminator to obtain the morphological category corresponding to the time series data segment. The morphological categories include impact type, oscillation type, and trend type. The diagnostic weights of conventional statistical features are determined based on the morphological category, including: pre-constructing a corresponding diagnostic feature weight vector for each morphological category, determining the diagnostic weights of conventional statistical features based on the diagnostic feature weight vector, calculating the adaptive weighted distance between the conventional statistical features of the current time series data segment and a variety of preset fault prototypes, and determining the final fault diagnosis result based on the fault prototype with the smallest adaptive weighted distance.

2. The method for fault diagnosis of automated storage and retrieval system equipment according to claim 1, characterized in that, The instantaneous impact index satisfies the following relationship: in, For instantaneous impact index, For the first time series data segment The value of each sampling point, This represents the total number of sampling points within the time-series data segment.

3. The method for fault diagnosis of automated storage and retrieval system equipment according to claim 1, characterized in that, The morphological skewness index satisfies the following relationship: in, This is the morphological skewness index. For the first time series data segment The value of each sampling point, This represents the total number of sampling points within the time-series data segment.

4. The method for fault diagnosis of automated storage and retrieval system equipment according to claim 1, characterized in that, The trend stability index satisfies the following relationship: in, As a trend stability index, , For symbolic functions, For the first time series data segment The value of each sampling point, This represents the total number of sampling points within a time-series data segment. This represents the total number of local trend changes within the entire time series data segment.

5. The method for fault diagnosis of automated storage and retrieval system equipment according to claim 1, characterized in that, The adaptive weighted distance satisfies the following relationship: in, This is the regular statistical feature vector of the current time series data segment. This is the fault prototype vector. The dimension of a conventional statistical feature vector. Based on morphological categories The determined diagnostic feature weight vector of the first The weights of each feature, and These are the first two elements in the real-time feature vector and the fault prototype vector, respectively. Each feature value.

6. The method for fault diagnosis of automated storage and retrieval system equipment according to claim 5, characterized in that, The conventional statistical characteristics include at least one of the following: mean, root mean square, variance, skewness, kurtosis, and peak value.

7. A fault diagnosis system for automated storage and retrieval system equipment, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a fault diagnosis method for an automated storage and retrieval system according to any one of claims 1-6.

Images