A fault early warning method and system for express logistics sorting equipment

Abstract

The present application belongs to the technical field of intelligent monitoring system of logistics equipment, and particularly relates to a fault early warning method and system for express logistics sorting equipment, which comprises the following steps: obtaining high-frequency vibration signals, identity information and real-time load weight of a sorting trolley; performing variational modal decomposition on the high-frequency vibration signals of the sorting trolley to obtain a plurality of modal components and extracting a time-domain feature vector; inputting the time-domain feature vector at the current time into an isolation forest model to calculate a bounded time series memory correction factor, and then calculating a time series memory anomaly score at the current time; and in response to the time series memory anomaly score being greater than an anomaly score threshold and the time rate of change of the time series memory anomaly score not being negative, triggering an early warning. The present application effectively eliminates noise and non-fault impact through adaptive modal screening, significantly reduces false alarms caused by incidental interference by using a time series memory enhancement mechanism, and realizes precise monitoring of the wear of the running wheels.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring systems for logistics equipment. More specifically, this invention relates to a method and system for early warning of faults in express logistics sorting equipment. Background Technology

[0002] With the automation upgrade of the logistics industry, cross-belt sorting machines have become core equipment in express delivery distribution centers and warehousing logistics systems. As the execution unit of the sorting machine, the sorting cart's wheels bear complex dynamic loads during long-term high-speed operation, making them prone to failures such as wheel wear. Once the wheels fail, it will not only lead to increased sorting noise and energy consumption, but may even cause the sorting cart to derail or jam, resulting in large-scale line stoppages and seriously affecting logistics timeliness.

[0003] Currently, the industry mainly uses non-contact vibration or sound sensors deployed alongside the sorting machine track for online monitoring. Existing signal processing and fault diagnosis techniques typically utilize variational mode decomposition to denoise the acquired vibration signals, and then input the extracted features into the isolated forest algorithm for anomaly detection.

[0004] However, the complex and variable environment of logistics sorting sites poses significant challenges to equipment condition monitoring. On the one hand, there is strong background noise, such as the operating noise of fans and chains; on the other hand, the load state of sorting carts is uncertain, varying randomly between empty and full load. Faced with such complex operating conditions, traditional variational mode decomposition methods typically rely on fixed energy percentages or correlation coefficient thresholds to filter modal components. This rigid screening mechanism is difficult to adapt to real-time fluctuations in load and noise, easily leading to distorted feature extraction. It may misjudge normal vibrations under heavy load as fault features, or mistakenly filter out genuine, weak wear features under light load as noise.

[0005] Furthermore, the traditional isolated forest algorithm only scores anomalies independently on data collected in a single instance, ignoring the gradual and cumulative evolution of wear and tear faults. Therefore, it is highly susceptible to accidental impacts on-site, such as cargo falling or impacts on track joints, which can cause scores to spike instantly and generate false alarms. This not only disrupts normal operation and maintenance processes but also increases the unnecessary burden of investigation on personnel. Summary of the Invention

[0006] To address the problems of severe distortion and extremely high false alarm rates in traditional variational mode decomposition methods under complex operating conditions, this invention provides a fault early warning method and system for express logistics sorting equipment, which can significantly improve the accuracy and robustness of fault early warning under complex operating conditions. This invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a fault early warning method for express logistics sorting equipment, comprising the following steps: acquiring the high-frequency vibration signal, identity information, and real-time load weight of the sorting cart; performing variational mode decomposition on the high-frequency vibration signal of the sorting cart to obtain several modal components, calculating the adaptive impact-correlation index of each modal component in combination with the real-time load weight, selecting the mode with the largest adaptive impact-correlation index as the optimal fault feature component and extracting the temporal feature vector; inputting the temporal feature vector at the current moment into an isolated forest model to obtain the average path length at the current moment, calculating the bounded temporal memory correction factor in combination with the anomaly score at the previous moment and the current feature deviation, and then calculating the temporal memory anomaly score at the current moment; triggering an early warning in response to the temporal memory anomaly score being greater than the anomaly score threshold and the time change rate of the temporal memory anomaly score not being negative.

[0008] By adopting the above technical solutions, synchronously collecting multi-source data and distinguishing between new and old vehicles to initialize baseline parameters provides an accurate data foundation and reference standard for subsequent feature extraction and anomaly scoring, thus solving the problem of monitoring new equipment due to a lack of historical data. Constructing an adaptive impact-correlation index that incorporates load correction and absolute value constraints effectively reduces the probability of statistical index failure under strong noise conditions and reduces the interference of load fluctuations on feature selection, ensuring the extraction of true fault features. Introducing a bounded temporal memory correction factor and a non-singular scoring model integrates historical state inertia and reduces the risk of computational overflow, enabling the scoring results to smoothly filter out occasional impacts while maintaining linear discrimination for severe faults. By combining scoring thresholds with a tiered early warning strategy based on trend confirmation, the false alarm rate can be further reduced, providing maintenance personnel with accurate early wear warnings and emergency shutdown protection, thereby optimizing maintenance plans.

[0009] Preferably, the formula for calculating the adaptive shock-correlation index is:

[0010]

[0011] in, Indicates the first One modal component; The validity score for the k-th modal component; The absolute value of the kurtosis value of the k-th modal component; It is a high-frequency vibration signal; For the first Modal components and high-frequency vibration signals Mutual information between them; To measure the real-time load weight of the sorting cart, The historical average load weight of the sorting cart; This is the load normalization reference constant; For adjustment coefficients, It is the hyperbolic tangent function.

[0012] By adopting the above technical solution, an adaptive impact-correlation index is constructed as the modal selection criterion. A load correction term based on the hyperbolic tangent function is introduced into the denominator. This correction term utilizes the boundedness of the hyperbolic tangent function to map the deviation between the real-time load weight and the historical average load weight to a finite interval [0,1). This not only achieves normalization of the load impact and effectively eliminates the distortion of feature evaluation indicators caused by drastic load fluctuations, such as heavy or light loads, but also avoids computational overflow or weight imbalance caused by excessive load deviation, ensuring that the algorithm can accurately lock the modal components containing fault information under full load conditions. Addressing the problem that traditional kurtosis indices are prone to negative values ​​in the complex, high-background-noise environment of the sorting site, leading to the failure of the evaluation system, this invention introduces the absolute value of the modal component kurtosis value in the numerator calculation and combines it with the mutual information index for comprehensive weighting. While ensuring fault impact sensitivity, it also considers the correlation between the decomposed components and the original signal, significantly enhancing the system's anti-interference capability and feature extraction robustness in harsh industrial environments. Preferably, the formula for the bounded temporal memory correction factor is:

[0013]

[0014] In the formula, Indicates the current moment; This is the temporal memory correction factor for the current moment; This is the memory decay coefficient; The final anomaly score for the previous moment; The feature vector at the current time; and These represent the mean and standard deviation of the characteristics of the sorting cart in a healthy state, respectively. To prevent division by zero errors for extremely small positive numbers.

[0015] By adopting the above technical solution and introducing a temporal memory correction factor, the limitation of the traditional isolated forest algorithm, which relies solely on isolated points at the current moment for judgment, is overcome. Integrating the fault state from the previous moment into the current score calculation allows the system to effectively smooth out data spikes caused by occasional instantaneous impacts such as track joint collisions and cargo drops, avoiding false alarms. When a real fault occurs, the system maintains the continuity and trend of the score, improving the confidence of fault identification. For extremely large abnormal feature values ​​that may arise in the late stage of a fault or sensor failure, this invention utilizes the saturation characteristics of the hyperbolic tangent function to construct a bounded constraint mechanism. Regardless of how much the current feature vector deviates from the healthy mean, the hyperbolic tangent function always maps the standardized feature deviation within the [0,1) interval. This design reduces the problem of correction factor calculation overflow or premature score saturation caused by feature value explosion, ensuring the numerical stability of the algorithm throughout its entire lifecycle monitoring.

[0016] Preferably, the formula for calculating the temporal memory abnormality score is:

[0017]

[0018] in, This is the final anomaly score; This represents the average path length of the current sample in the isolated forest. The average path length normalization constant for isolated forests depends on the sample size n. This is the temporal memory correction factor for the current moment; It is an exponential function with the natural constant as its base.

[0019] By adopting the above technical solution A dynamic, enhanced baseline was constructed, which is raised beyond the standard average path length when historical data shows an abnormal trend. Therefore, in the early or continuous stages of a fault, the system will automatically tighten the judgment criteria. Even if the path length of the current sample is only slightly shortened, the difference between the two will still increase significantly due to the rise of the baseline. Thus, the final anomaly score is extremely sensitive to continuous minor faults, effectively solving the problem that traditional algorithms have difficulty capturing gradual early wear.

[0020] When the system is in a healthy state Approaching zero, and with the average path length of the current sample in the isolated forest close to or greater than the standard average path length, a safety margin between isolated point determination and fault determination is defined by retaining a baseline coefficient of 1 in an exponential function with a natural constant as the base. Furthermore, it cleverly utilizes the mathematical boundary as a natural low-pass filter to automatically filter out high-frequency, short-term non-fault impacts, ensuring that the score only exceeds the safety margin and triggers an early warning when the anomaly has a cumulative effect over time.

[0021] Preferably, a fault early warning method for express logistics sorting equipment further includes calling historical benchmark parameters or performing cold start parameter initialization according to whether the sorting cart is a newly deployed vehicle. If the sorting cart is a newly deployed vehicle, the average benchmark value of the sorting cart group of the same model is called as the initial parameter, and only data is recorded without triggering alarms in the early stage of system operation. The benchmark parameters are updated using a sliding window statistical algorithm.

[0022] Preferably, the abnormal score threshold includes a warning threshold and a risk threshold. The step of triggering a warning in response to a time-series memory abnormal score being greater than the abnormal score threshold and the time change rate of the time-series memory abnormal score being non-negative includes: when the abnormal score exceeds the warning threshold, checking whether the abnormal scores of the most recent several periods maintain a non-decreasing trend; if the trend condition is met, it is determined to be early wear and a maintenance work order is generated; when the abnormal score exceeds the risk threshold, a shutdown protection command is directly triggered.

[0023] By setting early warning thresholds and risk thresholds, the operating status of sorting carts responding to time-series memory anomaly scores is classified, and different maintenance methods are executed according to the status of the carts. This reduces the probability of equipment safety accidents while improving operation and maintenance efficiency and economy.

[0024] Preferably, a fault early warning method for express logistics sorting equipment further includes preprocessing the high-frequency vibration signal after acquiring it: using the least squares method to linearly fit the vibration signal to obtain the optimal linear function, and subtracting the optimal linear function from the vibration signal to remove the trend term.

[0025] Preferably, selecting the mode with the largest adaptive shock-correlation index as the optimal fault feature component and extracting the time-domain feature vector includes: extracting the time-domain root mean square value and waveform factor of the optimal fault feature component, and constructing the time-domain feature vector by combining the parameters of the time-domain root mean square value and waveform factor.

[0026] Preferably, the abnormal scoring threshold includes a warning threshold and a risk threshold, which are determined by: selecting the 99th percentile of the abnormal scoring sample set of the sorting cart in its historical health state as the warning threshold, and adding a preset safety margin to the warning threshold to determine the risk threshold.

[0027] Secondly, the present invention also provides a fault early warning system for express logistics sorting equipment, 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 fault early warning method for express logistics sorting equipment is implemented.

[0028] By adopting the above technical solution, a computer program is generated from the above-mentioned fault early warning method for express logistics sorting equipment, and stored in the memory so that it can be loaded and executed by the processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0029] This invention creatively combines the impact characteristics of signals and information correlation by constructing an adaptive impact-correlation index, and introduces a load factor for adaptive adjustment. This enables the algorithm to accurately locate fault modes under noisy and variable load environments, solving the problems of feature extraction distortion and the easy submersion of effective information. At the same time, the temporal memory-enhanced isolated forest anomaly scoring mechanism proposed in this invention effectively distinguishes between occasional impacts and progressive wear by introducing historical score feedback and feature deviation correction. While ensuring high sensitivity to early weak faults, it significantly reduces the false alarm rate caused by operating condition fluctuations, realizing precise health management of the sorting cart throughout its entire life cycle and greatly improving operation and maintenance efficiency. Attached Figure Description

[0030] Figure 1 This is a flowchart of a fault early warning method for express logistics sorting equipment according to an embodiment of the present invention;

[0031] Figure 2 This is a comparison chart of the extraction effects of weak fault features under complex working conditions according to embodiments of the present invention;

[0032] Figure 3 This is a trend comparison chart of fault warning accuracy and anti-interference capability according to an embodiment of the present invention. Detailed Implementation

[0033] 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.

[0034] In a first aspect, embodiments of the present invention disclose a method for early warning of faults in express logistics sorting equipment.

[0035] Figure 1 This is a flowchart of a fault early warning method for express logistics sorting equipment according to an embodiment of the present invention, with reference to... Figure 1 It includes steps S1-S4:

[0036] S1. Obtain the high-frequency vibration signal, identification information, and real-time load weight of the sorting cart.

[0037] When acquiring the high-frequency vibration signal, identification information, and real-time load weight of the sorting cart, high-frequency vibration sensors and RFID radio frequency identification devices are first deployed at key monitoring points on the sorting machine track, such as curve entrances and pack loading points, in impact-sensitive areas. When the sorting cart passes through the monitoring area, the system performs the following operations:

[0038] 1. Use a high-frequency vibration sensor to synchronously acquire the vibration acceleration sequence at that moment. The sampling frequency is set to more than 5 times the characteristic frequency of wheel failure to ensure that the impact components caused by wear are captured.

[0039] 2. The vibration acceleration sequence at that moment was synchronously acquired using a high-frequency vibration sensor. Simultaneously, the RFID radio frequency identification device reads the tag on the sorting cart to obtain the cart's ID, and the real-time load weight of the sorting cart is obtained from the sorting control system. .

[0040] When it is necessary to obtain vehicle information, historical baseline parameters or cold start parameter initialization need to be performed depending on whether the vehicle is a newly deployed vehicle. Specifically, if the vehicle is already in operation, the average feature value of its health status is retrieved from the historical database. Standard deviation and historical average load If the vehicle is newly deployed, the average baseline value of a group of vehicles of the same model is used as the initial parameter. In the initial stage of system operation, only data recording is performed without triggering alarms. After accumulating sufficient health data, the baseline parameters are updated using a sliding window statistical algorithm, thereby solving the cold start monitoring problem for new equipment. It should be noted that in one embodiment of this example, the initial system operation is limited to the first 100 cycles of vehicle operation.

[0041] As a preferred option, the vibration acceleration sequence synchronously acquired by the high-frequency vibration sensor Subsequently, the collected vibration acceleration sequence should also be... Detrending processing is performed. The method for detrending the collected raw data is a mature technical solution in this field, with various mature methods available, such as: subtracting the mean, first-order polynomial fitting, higher-order polynomial fitting, moving average, or wavelet transform.

[0042] When monitoring the vibration of a sorting cart, the trend term is usually linearly drifting. Therefore, in this invention, it is preferable to use a first-order polynomial fitting method to fit the vibration acceleration sequence synchronously acquired by the high-frequency vibration sensor. The specific method for performing linear detrending processing is as follows:

[0043] 1. Based on the sampling time point sequence As the independent variable, with Using the least squares method as the dependent variable, the best linear function is obtained by fitting the data. ;

[0044] 2. Calculate the signal after detrending .

[0045] The vibration acceleration sequence synchronously acquired by the high-frequency vibration sensor After detrending, the signal is processed through a 4th-order Butterworth bandpass filter to remove extremely low-frequency structural vibrations and extremely high-frequency electromagnetic interference, while retaining the frequency band containing fault characteristics. The specific method is as follows:

[0046] 1. Calculate the fundamental frequency of the sorting cart's wheels based on the cart's travel speed and wheel diameter. ;

[0047] 2. Determine the passband range: Based on the impact response spectrum analysis of faults such as wheel wear and flat spots, the effective characteristic frequencies are mainly distributed between 2 and 50 times the wear level. Within this range. Therefore, the lower cutoff frequency of the 4th-order Butterworth bandpass filter is set. =2 Upper limit cutoff frequency =50 ;

[0048] 3. Perform filtering: The preprocessed vibration signal is obtained by using a fourth-order Butterworth bandpass filter designed with the above parameters, denoted as . .

[0049] Thus, by synchronously collecting and preprocessing vibration and load data, a clean and comprehensive data foundation is provided for subsequent accurate feature analysis under variable load conditions.

[0050] S2. Perform variational mode decomposition on the high-frequency vibration signal of the sorting trolley to obtain several modal components. Combine the real-time load weight to calculate the adaptive impact-correlation index of each modal component. Select the mode with the largest adaptive impact-correlation index as the optimal fault feature component and extract the time-domain feature vector.

[0051] In one implementation, to address the difficulty in extracting effective wear features against a noisy background, instead of directly using all the decomposed components, an evaluation index is constructed to screen the optimal mode.

[0052] When performing variational mode decomposition on the high-frequency vibration signal of the sorting cart to obtain several modal components, firstly, the preprocessed vibration signal... Perform variational mode decomposition to decompose it into Each intrinsic mode function component ,in Variational mode decomposition can effectively decompose complex multi-component signals into narrowband signals with specific center frequencies.

[0053] In one specific implementation, to adapt to the characteristics of the vibration signal of the sorting cart, the key parameters of variational mode decomposition are set as follows:

[0054] 1. Number of modes k: Estimated based on the number of independent spectral peaks in the main energy frequency band of the signal spectrum, where the main energy frequency band is the upper cutoff frequency of the sorting cart. The frequencies within the range correspond to the vibration of the walking wheel. Based on experience, the k value is set to a range of 6 to 10. The specific value of k can be determined by those skilled in the art by observing the distribution of the center frequencies of each mode after decomposition, and is not limited here.

[0055] 2. Penalty factor α: To ensure that the decomposed modes have sufficient frequency locality to separate fault features, while avoiding over-smoothing, the value of the penalty factor α needs to be determined based on the sampling frequency. As an implementation method, α can be set at the sampling rate. When the frequency is 10kHz, it is set to 2000. In practical applications, it can be fine-tuned around this reference value.

[0056] By setting the above parameters, variational mode decomposition can resolve composite signals that include noise, load fluctuations, and potential fault impacts. Decomposed into a series of narrowband modal components { , ,..., This lays the foundation for subsequent feature selection.

[0057] In order to This invention creatively constructs an adaptive impact-correlation index, denoted as [index name missing], to accurately identify modal components containing wheel wear impact characteristics from narrow-band modal components. It satisfies the following expression:

[0058]

[0059] in, This represents the validity score of the k-th modal component; It is the absolute value of the kurtosis value of the kth modal component. Taking its absolute value is to prevent negative kurtosis from causing exponential failure. It is the mutual information between the k-th modal component and the preprocessed vibration signal X; Indicates the first One modal component; This indicates the real-time load weight of the sorting cart. This indicates the historical average load weight of the sorting cart; It is the load normalization reference constant, which is usually the rated load of the trolley; For adjustment coefficients, It is a hyperbolic tangent function, which maps the load deviation term strictly to the interval [0,1).

[0060] Calculate all each modality The modal component with the largest value is selected as the optimal fault feature component. , and then from Extract features such as the root mean square value of time-domain statistics and waveform factors to construct a feature vector. .

[0061] The value reflects the impact characteristics of the modal components through kurtosis, and the information retention of the modal components and the original signal is calculated through mutual information. At the same time, the deviation between the real-time load and the average load is introduced. Finally, an adaptive adjustment term is formed by a nonlinear saturation function and an adjustment coefficient to suppress non-fault vibration interference caused by drastic load changes.

[0062] To explain more clearly The calculation process and function of the calculation are illustrated in a specific example provided in this embodiment of the invention:

[0063] Parameter setting details: Rated load capacity of the sorting cart Historical average load Adjustment coefficient The trolley is under heavy load and the real-time load is... For the first modal component obtained from the decomposition Its kurtosis value mutual information .

[0064] Example of calculation process:

[0065] Step 1: Calculate the numerator by introducing absolute value operations to handle negative kurtosis and ensure the non-negativity of the evaluation index:

[0066]

[0067] Step two, calculate the denominator. First, calculate the load deviation ratio and then use the hyperbolic tangent function for mapping:

[0068]

[0069]

[0070]

[0071] Step 3, calculate the final result value:

[0072]

[0073] The role of the adaptive shock-correlation index: Without using absolute values ​​and load correction, directly calculating kurtosis may result in negative scores, or excessive load deviations may lead to an overly large denominator, thus masking the true feature scores. This formula, even with large load fluctuations, limits the growth of the denominator to a reasonable range, and corrects negative kurtosis to positive values, ensuring the robustness of feature extraction. Thus, by calculating and filtering using the adaptive shock-correlation index, modal components containing fault information can be accurately identified under complex operating conditions, avoiding interference from noise and load fluctuations.

[0074] S3. Input the temporal feature vector of the current moment into the isolated forest model to obtain the average path length of the current moment. Combine the anomaly score of the previous moment with the current feature deviation to calculate the bounded temporal memory correction factor, and then calculate the temporal memory anomaly score of the current moment.

[0075] First, the feature vector at the current time. Input a pre-trained isolation forest model. For this input sample, calculate its average path length across all isolation trees. ,in, That is to say .in, The smaller the value, the shorter the path length of the sample in the preset isolated forest model at the current moment, and the easier it is for the sample to be isolated, that is, the more likely the sample is to be an outlier.

[0076] It should be noted that the preset isolated forest model should be an isolated forest model trained using the health data of the sorting carts. The health data of the sorting carts is data collected from carts that have been confirmed to be fault-free through manual or long-term monitoring. The training method of the preset isolated forest model is a method already existing in the art; however, for ease of understanding, this embodiment provides a brief supplementary explanation of the construction method of the preset isolated forest model:

[0077] 1. Training data preparation: Collect a large number of vibration signal samples of the sorting cart under historical healthy operating conditions. After processing through the above steps, a training dataset consisting of health status feature vectors is obtained.

[0078] 2. Model Parameter Settings: Based on the size and characteristics of the training dataset, set the number of trees in the preset isolated forest model and the subsampling size of each tree. As one implementation, the number of trees in the preset isolated forest model can be set to 100, and the subsampling size of each tree can be set to 256. If the total number of samples in the training dataset is less than 256, all samples are used. The number of trees and the subsampling size of each tree can also be adjusted by those skilled in the art in actual operation according to the actual situation, and there are no restrictions here.

[0079] 3. Model Training: A preset isolated forest model is generated based on the standard isolated forest algorithm. After training, the preset isolated forest model is fixed and used for path length calculation during online monitoring.

[0080] To address the issue that traditional isolated forests ignore fault evolution trends, this invention introduces a temporal memory correction mechanism in anomaly scoring calculation to reduce the probability of false alarms caused by environmental interference and improve the robustness of the system.

[0081] To utilize the temporal continuity of fault evolution to suppress sporadic interference, this invention constructs a bounded temporal memory correction factor. It satisfies the following expression:

[0082]

[0083] in, Indicates the current moment; Represents the bounded temporal memory correction factor at the current moment; This indicates the abnormal score from the previous moment; For memory weighting coefficients; This represents the feature vector value at the current moment; and These are the mean and standard deviation of the feature vector of the sorting cart under its historical health status, respectively; It is a very small positive number.

[0084] After obtaining the bounded temporal memory correction factor Next, calculate the temporal memory anomaly score at the current moment. Its calculation formula adopts a non-singular scoring formula based on a Sigmoid variant, specifically:

[0085]

[0086] in, It is the final anomaly score at the current moment; It is the normalization constant for the average path length of the isolated forest, which depends on the sample size n.

[0087] The following is a specific example to illustrate how the temporal memory mechanism suppresses false alarms. The scenario is as follows: the wheels of the sorting cart are in a stage of significant wear, and the anomaly score from the previous moment was high. The system has already acquired fault inertia through temporal memory.

[0088] Parameter configuration details: Isolation forest normalization constant The current sample is identified as an anomaly by the isolated forest because of its short average path length. Previous time-lapse score Memory decay coefficient Since the current feature deviation is large, we assume the standardized deviation... .

[0089] Example of calculation process:

[0090] Step 1: Calculate the bounded temporal memory correction factor The saturation property of the hyperbolic tangent function is used to handle feature deviation, combined with historical scores:

[0091]

[0092]

[0093] Step 2: Calculate the index term parameters in the scoring model, and calculate the path length deviation after introducing the dynamic correction factor:

[0094]

[0095]

[0096] Step 3: Calculate the final anomaly score :

[0097]

[0098] Results verification: The final score of 0.768 exceeded the general warning threshold and accurately reflected the risk of failure.

[0099] Stability Comparison: Without this formula, large feature deviations may lead to numerical instability in traditional algorithms, while in the embodiments of this invention, even... When the theoretical maximum value of 1 is reached, the exponent term is still a finite value, which ensures that the score will only smoothly approach 1 without causing computational overflow, thus guaranteeing the numerical stability of the system under extreme failure conditions.

[0100] S4. If the abnormal score of the temporal memory is greater than the abnormal score threshold and the time change rate of the abnormal score of the temporal memory is not negative, an early warning will be triggered.

[0101] As one implementation method, the setting of anomaly scoring thresholds may include warning thresholds and risk thresholds:

[0102] The warning threshold can be set to the 99th percentile of the health score sample set. This setting means that when the car is in a healthy state, its abnormal score has only about a 1% probability of exceeding this threshold due to random fluctuations, thus laying the foundation for a low false alarm rate at the algorithm level.

[0103] The risk threshold is set to a fixed value that is much higher than the warning threshold to ensure that serious faults can be clearly identified.

[0104] Based on the comparison results, the status of the sorting cart's wheels is determined, and corresponding warning operations are executed, including:

[0105] If the final anomaly score is less than the preset warning threshold, the sorting cart is determined to be in a normal state or only subject to occasional interference. The system will not issue an alarm, but will only record the data.

[0106] If the final anomaly score exceeds the warning threshold multiple times consecutively and shows a monotonous upward trend, it is determined that the walking wheels have experienced early wear. The system automatically generates a low-level warning work order suggesting inspection, prompting maintenance personnel to pay attention to the sorting cart during the next downtime maintenance window.

[0107] If the final abnormal score exceeds the risk threshold, and is accompanied by A sharp increase in the value indicates a serious malfunction. The system immediately triggers an audible and visual alarm and sends a command to the control system to automatically guide the sorting cart to the maintenance area or recommend an emergency shutdown.

[0108] In one embodiment, the methods for determining the warning threshold, the number of times the final anomaly score exceeds the warning threshold consecutively, and the risk threshold are as follows:

[0109] Warning threshold: Based on the score distribution of historical health datasets, the 99th percentile of the health score sample set is selected. In a specific embodiment, this value is set to 0.6.

[0110] The number of times the final anomaly score exceeds the warning threshold consecutively: determined based on the system's sampling interval and the timescale of early wear development, preferably 3 to 5 times, and in one specific embodiment, this value is set to 3 times;

[0111] Risk threshold: A safety margin is added to the warning threshold. The safety margin is set according to the acceptable risk of false negatives. In a specific embodiment, the safety margin is set to 0.25, and the risk threshold is set to 0.85.

[0112] The specific values ​​of the warning threshold, the number of times the final abnormal score exceeds the warning threshold consecutively, and the risk threshold can be determined by the staff based on the actual operation of the sorting cart, and are not limited here.

[0113] In summary, through the aforementioned hierarchical early warning mechanism based on the final anomaly score and adaptive impact-correlation index, the system can intelligently distinguish between three different states: occasional interference, early wear, and severe failures, and trigger differentiated responses ranging from data recording and planned inspections to emergency intervention. This not only reduces the probability of equipment safety accidents caused by unreported severe failures but also reduces the probability of unplanned downtime through accurate early warnings, realizing a shift from reactive maintenance to predictive maintenance and significantly improving operational efficiency and economy.

[0114] like Figure 2 As shown in the figure, this comparison demonstrates the effectiveness of the present invention in extracting weak fault features under complex working conditions. The original vibration signal collected on-site contained both strong background noise and high-amplitude sudden interference in the t=200 to 250 interval. After processing in step S2 of the present invention, the reconstructed feature signal in the sudden interference region on the left has a curve amplitude close to 0, indicating that the present invention has successfully identified and eliminated the non-fault impact component. In the region on the right with t>500, the reconstructed feature signal curve clearly shows periodic fault pulses as the wear degree increases, and the pulse amplitude gradually increases. This proves that the present invention has the ability to effectively extract weak early wear features even in a strong noise background.

[0115] like Figure 3 As shown in the figure, this graph compares the accuracy of fault warning and the system's anti-interference capability. The anomaly score curve of the prior art rises sharply during the period of occasional interference on the left, triggering false alarms, while the anomaly score of the present invention only fluctuates slightly and remains below the warning threshold. During the fault development stage on the right, the anomaly score of the present invention shows a smooth and monotonically rising trend, accurately crossing the warning threshold, thus achieving precise and timely warning.

[0116] Secondly, embodiments of the present invention also disclose a fault early warning system for express logistics sorting equipment, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned fault early warning method for express logistics sorting equipment is implemented.

[0117] 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.

[0118] 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.

Claims

1. A method for early warning of faults in express logistics sorting equipment, characterized in that, include: The high-frequency vibration signal, identity information, and real-time load weight of the sorting cart are obtained. After obtaining the high-frequency vibration signal of the sorting cart, the high-frequency vibration signal is preprocessed: the least squares method is used to linearly fit the high-frequency vibration signal to obtain the best linear function, and the function value of the best linear function at the same time is subtracted from the vibration signal to remove the trend term. The signal after the detrending term is processed through a fourth-order Butterworth bandpass filter to filter out extremely low-frequency structural vibrations and extremely high-frequency electromagnetic interference, while retaining frequency bands containing fault characteristics, including: calculating the fundamental frequency of the sorting cart's wheels based on the cart's travel speed and wheel diameter. Based on the impact response spectrum analysis of wheel wear and flat spot failures, the effective characteristic frequencies are mainly distributed between 2 and 50 times. Within the specified range, set the lower cutoff frequency of the 4th-order Butterworth bandpass filter. =2 Upper limit cutoff frequency =50 The de-stressed signal is passed through a fourth-order Butterworth bandpass filter designed with the above parameters to obtain the pre-processed vibration signal. Variational mode decomposition is performed on the high-frequency vibration signal of the sorting cart to obtain several modal components. The adaptive shock-correlation index of each modal component is calculated based on the real-time load weight. The mode with the largest adaptive shock-correlation index is selected as the optimal fault feature component, and its time-domain feature vector is extracted. The adaptive shock-correlation index satisfies: Indicates the first One modal component; The validity score for the k-th modal component; The absolute value of the kurtosis value of the k-th modal component; It is a high-frequency vibration signal; For the first Modal components and high-frequency vibration signals Mutual information between them; To measure the real-time load weight of the sorting cart, The historical average load weight of the sorting cart; This is the load normalization reference constant; For adjustment coefficients, It is the hyperbolic tangent function; The current temporal feature vector is input into the isolated forest model to obtain the average path length at the current time. The bounded temporal memory correction factor is calculated by combining the anomaly score of the previous time step with the current feature deviation, and then the temporal memory anomaly score at the current time step is calculated. An alert is triggered if the time-series memory anomaly score is greater than the anomaly score threshold and the time change rate of the time-series memory anomaly score is not negative.

2. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, The formula for the bounded temporal memory correction factor is: In the formula, Indicates the current moment; This is the temporal memory correction factor for the current moment; This is the memory decay coefficient; The final anomaly score for the previous moment; The feature vector at the current time; and These represent the mean and standard deviation of the characteristics of the sorting cart in a healthy state, respectively. To prevent division by zero errors for extremely small positive numbers.

3. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, The formula for calculating the temporal memory abnormality score is as follows: in, This is the final anomaly score; This represents the average path length of the current sample in the isolated forest. The average path length normalization constant for isolated forests depends on the sample size n. This is the temporal memory correction factor for the current moment; It is an exponential function with the natural constant as its base.

4. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, Also includes: Depending on whether the sorting cart is a newly deployed vehicle, historical baseline parameters or cold start parameters are initialized. If the sorting cart is a newly deployed vehicle, the average baseline value of the same model of sorting carts is used as the initial parameter. In the early stage of system operation, only data is recorded and no alarm is triggered. The baseline parameters are updated using a sliding window statistical algorithm.

5. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, The abnormal scoring threshold includes a warning threshold and a risk threshold. The step of triggering a warning in response to the time-series memory abnormal score being greater than the abnormal scoring threshold and the time change rate of the time-series memory abnormal score being non-negative includes: when the abnormal score exceeds the warning threshold, checking whether the abnormal scores of the most recent several periods maintain a non-decreasing trend; if the trend condition is met, it is determined to be early wear and a maintenance work order is generated; when the abnormal score exceeds the risk threshold, a shutdown protection command is directly triggered.

6. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, Selecting the mode with the largest adaptive shock-correlation index as the optimal fault feature component and extracting the time-domain feature vector includes: Extract the root mean square value of the time domain statistics and the waveform factor of the optimal fault feature component, and construct the time domain feature vector by combining the parameters of the root mean square value of the time domain statistics and the waveform factor.

7. The method for early warning of faults in express logistics sorting equipment according to claim 1, characterized in that, The abnormal scoring threshold includes a warning threshold and a risk threshold. The warning threshold is determined by selecting the 99th percentile of the abnormal scoring sample set of the sorting cart in its historical health state based on the abnormal scoring sample set. A preset safety margin is added to the warning threshold to determine the risk threshold.

8. A fault early warning system for express logistics sorting equipment, comprising a processor and a memory, characterized in that, The memory stores computer program instructions, which, when executed by the processor, implement a fault early warning method for express logistics sorting equipment according to any one of claims 1-7.

Images