Method and system for detecting abnormal operation of mining excavator

Abstract

The invention relates to the technical field of industrial machinery monitoring, in particular to a mining excavator operation anomaly detection method and system. The method comprises the steps that firstly, operation data of the excavator are obtained, and the current operation working condition of the excavator in a set of predefined operation working conditions is determined through the operation data; then, a dynamic statistical base line corresponding to the current working condition is selected from a plurality of stored dynamic statistical base lines, and each base line represents the normal hydraulic system behavior under the specific working condition; and finally, determining whether an abnormal condition exists in the hydraulic system or not by calculating the statistical deviation degree between the real-time hydraulic data of the excavator and the selected dynamic statistical base line. By dynamically associating the anomaly detection standard with the specific operation condition, the accuracy and reliability of detection are remarkably improved, and the false alarm rate generated in complex dynamic operation by a traditional method is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial machinery monitoring, in particular to a mining excavator operation anomaly detection method and system. BACKGROUND

[0002] The mining excavator is the core equipment in mine exploitation and large earthwork engineering. Its hydraulic system, as the key part of power transmission and control, operates in high pressure, heavy load and variable harsh environment for a long time. The leakage of hydraulic oil is one of the most common faults of the hydraulic system. The slight leakage not only causes economic loss and environmental pollution, but also may evolve into serious faults such as insufficient system pressure and actuator malfunction, and even cause disastrous equipment damage and safety accidents. Therefore, real-time and accurate anomaly detection of the excavator hydraulic system, especially early warning of the slight leakage, is of great significance to ensure equipment safety, improve production efficiency and protect the environment.

[0003] The traditional leakage detection scheme usually relies on direct measurement of physical parameters of the hydraulic system and static threshold judgment. This method is simple in principle, but in the dynamic environment of actual operation of the excavator, the "noise" fluctuation caused by normal operation is much larger than the slow decline of the liquid level caused by early slight leakage, so that the leakage signal is completely submerged. Therefore, this kind of method can only be effective when the equipment is in a static or near-static state, and cannot provide effective protection in the key operation stage when the equipment is most active and the hydraulic system bears the maximum pressure.

[0004] The fault diagnosis method based on signal processing and machine learning analyzes multi-dimensional sensor signals such as vibration and pressure, and uses algorithms such as support vector machine and neural network to identify fault patterns. Although these studies prove the feasibility of using sensor data for fault diagnosis, they usually model the hydraulic system as a whole, which is complex, difficult to train and interpret. SUMMARY

[0005] In order to overcome the above-mentioned defects in the prior art, the present application provides a mining excavator operation anomaly detection method and system.

[0006] In one aspect, the present application provides a mining excavator operation anomaly detection method, comprising the following steps: Using the operation sensor deployed on the mining excavator, operation data of the mining excavator is obtained; Based on the operation data, one or more features are extracted to form a feature vector; Based on the feature vector, a pre-trained operation condition classifier is used to identify the current operation condition of the mining excavator; acquiring hydraulic data of the mining excavator by using a hydraulic sensor arranged on the mining excavator; based on the current working condition of the mining excavator, calling a corresponding dynamic statistical baseline, the dynamic statistical baseline being used to define a normal operation mode of the hydraulic system of the mining excavator under the working condition; quantifying a statistical deviation between the hydraulic data of the mining excavator and the dynamic statistical baseline; based on the statistical deviation, determining an abnormal condition of the hydraulic system of the mining excavator.

[0007] Preferably, the training method of the working condition classifier specifically comprises: synchronously collecting time series data including operation data and video data recording operation actions of one or more mining excavators during operation of the mining excavators; dividing the time series data into a plurality of data segments with a predetermined time length; based on analysis of the video data synchronized with the data segments, assigning a working condition label selected from a set of predefined working condition categories to each data segment as the working condition of the data segment; extracting a multi-dimensional feature vector for each labeled data segment; adding the working condition of the data segment and the multi-dimensional feature vector corresponding thereto to a data set; training a machine learning model by using the labeled data set containing a plurality of the feature vectors and their corresponding working condition labels to generate the working condition classifier.

[0008] Preferably, the method for establishing the dynamic statistical baseline specifically comprises: collecting multi-dimensional hydraulic data of a mining excavator confirmed to be in a fault-free state when operating under a plurality of predefined working conditions; grouping the collected multi-dimensional hydraulic data according to their corresponding working conditions; for each working condition group, performing data fusion on the grouped multi-dimensional hydraulic data to construct a health feature vector representing a normal healthy state under the working condition; based on the health feature vector, calculating the mean vector and the covariance matrix by statistical analysis as the dynamic statistical baseline corresponding to the working condition.

[0009] Preferably, the data fusion is realized by a principal component analysis method, and the health feature vector is composed of the first three principal components of the multi-dimensional hydraulic data.

[0010] Preferably, the dynamic statistical baseline is a multivariate statistical model, specifically a multi-dimensional Gaussian distribution defined by a mean vector and a covariance matrix.

[0011] Preferably, the determination of the dynamic statistical baseline corresponding to the working condition of the operation based on the health feature vector specifically includes: calculating a sample mean vector of the health feature vector for the working condition group; calculating a sample covariance matrix of the health feature vector for the working condition group; The sample mean vector and the sample covariance matrix jointly constitute the dynamic statistical baseline.

[0012] Preferably, the quantification of the statistical deviation of the hydraulic data of the mining excavator and the dynamic statistical baseline is specifically quantified by calculating the Mahalanobis distance of the hydraulic data of the mining excavator and the dynamic statistical baseline, and the calculation formula is: ; In the formula, is the Mahalanobis distance of the hydraulic data of the mining excavator and the dynamic statistical baseline, is a vector composed of the hydraulic data of the mining excavator, is a mean vector of the dynamic statistical baseline, is a covariance matrix of the dynamic statistical baseline.

[0013] Preferably, the quantification of the statistical deviation of the hydraulic data of the mining excavator and the dynamic statistical baseline further includes further calculation by applying an exponential weighted moving average algorithm to a time series of the Mahalanobis distance.

[0014] Preferably, the determination of the abnormal condition of the hydraulic system of the mining excavator based on the statistical deviation is specifically: when the statistical deviation exceeds a threshold value preset for the current working condition, it is determined that the hydraulic system is in an abnormal state.

[0015] Another aspect of the present application also provides a mining excavator operation anomaly detection system, comprising a data acquisition module, an operation condition determination module, a dynamic baseline management module, an anomaly detection module and a warning module. The data acquisition module is used to collect analog or digital signals from the operation data sensor and the hydraulic data sensor, and transmit the data to the subsequent modules after preprocessing. The operation condition determination module is used to determine the current working condition of the mining excavator according to the operation data collected by the data acquisition module. The dynamic baseline management module is used to select a dynamic statistical baseline according to the current working condition of the mining excavator. The abnormality detection module is configured to determine the abnormal state of the hydraulic system of the mining excavator according to the statistical deviation of the hydraulic data collected by the data collection module and the dynamic statistical baseline. The early warning module is configured to generate and issue a warning when it is determined that the hydraulic system is in an abnormal state.

[0016] Compared with the prior art, the present application has the following advantages: By first accurately identifying the current working condition of the excavator, and then calling the dynamic statistical baseline strictly matched with the working condition for comparison, the present application converts a complex, multi-modal abnormality detection problem into a series of simple, single-modal detection problems. This context-based adaptive detection mechanism can effectively filter out "false abnormality" signals caused by normal working condition switching, greatly improving the detection sensitivity and accuracy of real abnormalities (such as early hydraulic oil leakage), while significantly reducing the false positive rate, providing a reliable technical guarantee for the safe operation and predictive maintenance of the mining excavator. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A method flowchart of a mining excavator operation abnormality detection method according to the present application is shown in the figure. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] REFERENCE Figure 1 A mining excavator operation abnormality detection method according to the present application is described. Specifically, it includes: S1, using the operation sensors deployed on the mining excavator to obtain the operation data of the mining excavator, and using the AI camera to obtain the bucket tooth recognition state data of the mining excavator to detect the bucket tooth recognition abnormal state.

[0020] The sensors can include a vehicle-mounted bus interface, an inertial measurement unit, and a GPS receiver, and also include a bucket tooth detection AI camera to detect the bucket tooth abnormality (bucket tooth falling off) recognition function, and an alarm is issued when the bucket tooth falling off is recognized.

[0021] The vehicle bus interface is used to connect to the controller area network of the excavator, and directly read various parameters broadcasted by the vehicle electronic control unit, such as engine speed, throttle pedal opening, command and actual pressure of the main pump and pilot pump, stroke or angle of each hydraulic actuator (such as boom, stick, bucket cylinder), speed and torque of the swing motor, and position signal of the operating handle, etc.

[0022] The inertial measurement unit (IMU) is used to measure the three-axis acceleration and three-axis angular velocity of the component it is located in.

[0023] The GPS receiver is used to obtain the geographical position, travel speed and trajectory of the excavator.

[0024] In an embodiment, the operation data includes: engine speed, command and actual pressure of the main pump and pilot pump, stroke or angle of each hydraulic actuator (including but not limited to: boom, stick, bucket cylinder), speed and torque of the swing motor.

[0025] S2, based on the operation data, extracting one or more features to constitute a feature vector.

[0026] It can be understood that the original, high-frequency time series sensor data cannot be directly used by machine learning models. It must go through a series of preprocessing and feature engineering steps to be transformed into structured data that can effectively represent the running state of the excavator in a short period of time, i.e. a feature vector. Specifically: First, the received continuous multi-channel data stream is segmented. In a preferred embodiment, a fixed length sliding window method is used. For example, a 5 second window length is selected, and a 50% overlap rate is set.

[0027] Then, for each 5 second data window, a series of numerical features are calculated, and the dynamic information in this window is compressed into a fixed dimension feature vector. In an embodiment, the extracted features include statistical features, frequency domain features and cross-correlation features.

[0028] Statistical features: These features describe the basic distribution characteristics of the signal within the window. For example, for channels such as engine speed, command and actual pressure of the main pump and pilot pump, calculate their mean, variance, standard deviation, root mean square, kurtosis and skewness. The mean and root mean square reflect the average energy or load level of the signal, the variance and standard deviation reflect its volatility, and the kurtosis and skewness describe the shape of the signal distribution, which can capture subtle changes such as shocks or asymmetry.

[0029] Frequency domain features: For vibration and motion signals (e.g., acceleration and angular velocity data of hydraulic actuators and rotary motors measured by IMU), perform fast Fourier transform to analyze their frequency components. Extract features such as dominant frequency, spectral entropy, etc. from the frequency spectrum. For example, the hydraulic system may produce specific high-frequency noise when there is an abnormal leak, which is not obvious in the time domain but clear in the frequency domain.

[0030] Cross-correlation features: Calculate the correlation features between different signal channels through cross-correlation function. For example, calculate the cross-correlation coefficient between the boom cylinder pressure and the stick cylinder pressure.

[0031] Finally, combine the above calculated features into a single, high-dimensional feature vector.

[0032] S3, based on the feature vector, use a pre-trained working condition classifier to identify the current working condition of the mining excavator.

[0033] Specifically: input the feature vector into a pre-trained working condition classifier, and the working condition classifier outputs the current working condition of the mining machine.

[0034] Wherein, the training method of the working condition classifier is: First, select one or more representative excavators (e.g., different tonnage, different manufacturer models) to collect data in their real working environment (e.g., mines, quarries), and simultaneously record two types of data streams: Multi-channel sensor data: The same as the data collected in S1, ensuring that the training data and the actual deployment data have consistent format and source.

[0035] Synchronous video data: Install one or more cameras in the cab of the excavator or externally to clearly record each action of the excavator's working device and its interaction with the environment.

[0036] It should be noted that the video stream must be accurately time-synchronized with the sensor data stream, for example, through the Network Time Protocol or hardware trigger signals.

[0037] After that, the working condition annotation is done by experienced excavator operators or mechanical engineers: first, define a clear, complete and mutually exclusive set of working condition categories. For example: idle, walking, positioning, digging, load swing and unloading, etc. Then, play the video and sensor data chart simultaneously, and the annotator will label each 5-second sensor data segment with a working condition label that best describes the main activity in that time period according to the observed action of the excavator in the video. For example, when the video shows the bucket cutting into the earthwork, the corresponding 5-second data segment is marked as "digging". This step will produce a high-quality labeled data set, in which each data sample contains a sensor data segment and a verified "correct answer" (i.e. working condition label) corresponding to it.

[0038] Next, for each 5-second data segment in the labeled data set, apply the feature extraction method described in S2 to convert it into a feature vector. After this step, the original labeled data set is converted into a structured data set consisting of (feature vector, working condition label) pairs, which can be directly used for training machine learning models.

[0039] Next, select a supervised learning classification algorithm (such as the gradient boosting decision tree model) to train the working condition classifier: input the (feature vector, working condition label) data set generated in the above steps into the gradient boosting decision tree algorithm; the algorithm iteratively builds a series of decision trees, each new tree aims to correct the prediction errors of the previous tree, and finally combines into a powerful ensemble model. The model has learned the complex mapping relationship from the input sensor feature combination to the specific working condition.

[0040] Finally, use this model as the working condition classifier in this scheme.

[0041] S4, obtain hydraulic data of the mining excavator using hydraulic sensors deployed on the mining excavator.

[0042] The hydraulic data sensor can include a pressure sensor, a temperature sensor and a flow meter.

[0043] The pressure sensor is installed at key nodes of the hydraulic system, such as the main pump outlet, the oil inlet and return pipeline of each actuator cylinder, and the pilot control oil circuit, for real-time monitoring of system pressure.

[0044] The temperature sensor is installed in the hydraulic oil tank or main return oil pipeline to monitor the temperature of the hydraulic oil.

[0045] The flow meter is installed on some key pipelines (such as the main pump outlet or return oil circuit) to directly measure the flow of hydraulic oil.

[0046] In an embodiment, the acquired hydraulic data of the mining excavator includes: main pump outlet pressure, oil inlet pressure and return line pressure of each actuator cylinder, pilot control oil line pressure, hydraulic oil temperature and hydraulic oil flow.

[0047] S5, calling a corresponding dynamic statistical baseline based on the current working condition of the mining excavator.

[0048] The dynamic statistical baseline defines the normal operation mode of the hydraulic system of the mining excavator under the working condition.

[0049] The method for establishing the dynamic statistical baseline is: First, on a mining excavator confirmed to be fault-free, make it perform standard operations under various working conditions. During this period, the data acquisition module synchronously acquires the same data as acquired in S1, and at the same time, labels each piece of data with a corresponding working condition.

[0050] Then, group the acquired multi-dimensional hydraulic data according to the working condition labels. In order to more effectively represent the system health state and reduce the data dimension, data fusion can be performed on the multi-dimensional hydraulic data in each working condition group to construct a "health feature vector". In a preferred embodiment, the data fusion is realized through principal component analysis (PCA). PCA can recombine the original, possibly correlated hydraulic parameters into a new set of uncorrelated variables (i.e. principal components), and select the first few principal components that can explain most of the data variance to form the health feature vector.

[0051] Finally, for the health feature vector of each working condition group, a multivariate statistical model is established, which is the dynamic statistical baseline of the working condition.

[0052] In a preferred embodiment, the statistical model is a multi-dimensional Gaussian distribution. For a working condition The baseline is defined by a mean vector and a covariance matrix The mean vector represents the average expected value of each health feature under the working condition, and the covariance matrix captures the variance (i.e. the range of normal fluctuations) of these features and their correlation.

[0053] S6, quantifying the statistical deviation of the hydraulic data of the mining excavator and the dynamic statistical baseline.

[0054] In this scheme, the statistical deviation is quantified by calculating the Mahalanobis distance between the hydraulic data of the mining excavator and the dynamic statistical baseline. The calculation formula is: ; In the formula, Mahalanobis distance of the hydraulic data and dynamic statistical baseline of the mining excavator, vector composed of the hydraulic data of the mining excavator, mean vector of the dynamic statistical baseline, covariance matrix of the dynamic statistical baseline.

[0055] In a preferred embodiment, in order to smooth transient fluctuations and improve the sensitivity to persistent minor anomalies (such as slow leaks), an Exponentially Weighted Moving Average (EWMA) algorithm is applied to the time series of Mahalanobis distance. At each time step , the system updates the EWMA statistic as the final statistical deviation: ; where is the smoothing constant, is the Mahalanobis distance at the current time step, is the EWMA value at the previous time step.

[0056] S7, based on the statistical deviation, determining the abnormal condition of the hydraulic system of the mining excavator.

[0057] Specifically, the calculated EWMA statistical deviation is compared with the threshold value preset for the current working condition.

[0058] If , it is determined that the current hydraulic system state is normal, and the process returns to continue monitoring.

[0059] If , it is determined that an abnormal event is detected.

[0060] Once an abnormal state is determined, the warning module generates an alarm signal and sends an alarm to the operator or the background monitoring center through the visualization interface. The alarm information can include the current working condition, the excessive deviation and the possible related hydraulic parameters, providing clues for subsequent troubleshooting and maintenance.

[0061] In order to realize the above-mentioned abnormal detection method for the hydraulic system of the mining excavator, the present application further provides a mining excavator operation abnormal detection system, comprising a data acquisition module, an operation condition determination module, a dynamic baseline management module, an abnormal detection module and a warning module.

[0062] The data acquisition module is used to collect analog or digital signals from operation data sensors and hydraulic data sensors, and transmit the data to subsequent modules after pretreatment.

[0063] ​The working condition determination module is configured to determine the current working condition of the mining excavator according to the operation data collected by the data collection module.

[0064] The dynamic baseline management module is configured to select a dynamic statistical baseline according to the current working condition of the mining excavator.

[0065] The abnormality detection module is configured to determine the abnormal state of the hydraulic system of the mining excavator according to the statistical deviation of the dynamic statistical baseline from the hydraulic data collected by the data collection module.

[0066] The early warning module is configured to generate and issue a warning when it is determined that the hydraulic system is in an abnormal state.

[0067] In the description of the present specification, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0068] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of the present specification. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. A method of detecting abnormality of operation of a mining excavator, characterized by, The method comprises the following steps: obtaining operation data of the mining excavator by using operation sensors deployed on the mining excavator; extracting one or more features based on the operation data to form a feature vector; identifying a current working condition of the mining excavator by using a pre-trained working condition classifier based on the feature vector; obtaining hydraulic data of the mining excavator by using hydraulic sensors deployed on the mining excavator; calling a corresponding dynamic statistical baseline based on the current working condition of the mining excavator, the dynamic statistical baseline being used to define a normal operation mode of the hydraulic system of the mining excavator under the working condition; quantifying a statistical deviation of the hydraulic data of the mining excavator and the dynamic statistical baseline; determining an abnormal condition of the hydraulic system of the mining excavator based on the statistical deviation.

2. The method of claim 1, wherein The training method of the working condition classifier comprises the following steps: synchronously collecting time series data including operation data and video data recording operation actions of one or more mining excavators during operation of the mining excavators; dividing the time series data into a plurality of data segments with a predetermined time length; assigning a working condition label selected from a set of predefined working condition categories to each data segment as the working condition of the data segment based on analysis of video data synchronized with the data segment; extracting a multi-dimensional feature vector for each labeled data segment; adding the working condition of the data segment and the multi-dimensional feature vector corresponding to the working condition to a data set; training a machine learning model by using a labeled data set containing a plurality of feature vectors and corresponding working condition labels to generate the working condition classifier.

3. The method of claim 1, wherein The establishment method of the dynamic statistical baseline comprises the following steps: collecting multi-dimensional hydraulic data of a confirmed fault-free mining excavator when operating under a plurality of predefined working conditions; grouping the collected multi-dimensional hydraulic data according to corresponding working conditions; performing data fusion on the grouped multi-dimensional hydraulic data to construct a health feature vector representing a normal and healthy state under the working condition for each working condition group; determining a dynamic statistical baseline corresponding to the working condition based on the health feature vector.

4. The method of claim 3, wherein The data fusion is realized by a principal component analysis method, and the health feature vector is composed of the first three principal components of the multi-dimensional hydraulic data.

5. The method of claim 3, wherein the method further comprises: The dynamic statistical baseline is a multivariate statistical model, specifically a multi-dimensional Gaussian distribution defined by a mean vector and a covariance matrix.

6. The method of claim 3, wherein the method further comprises: The determination of the dynamic statistical baseline corresponding to the working condition based on the health feature vector comprises the following steps: calculating a sample mean vector of the health feature vector for the working condition group; calculating a sample covariance matrix of the health feature vector for the working condition group; the sample mean vector and the sample covariance matrix jointly constitute the dynamic statistical baseline.

7. The method of claim 1, wherein the method further comprises: The quantification of the statistical deviation of the hydraulic data of the mining excavator and the dynamic statistical baseline is quantified by calculating the Mahalanobis distance of the hydraulic data of the mining excavator and the dynamic statistical baseline, and the calculation formula is: ； wherein is the Mahalanobis distance of the hydraulic data and dynamic statistical baseline of the mining excavator, is the vector of the hydraulic data of the mining excavator, is the mean vector of the dynamic statistical baseline, is the covariance matrix of the dynamic statistical baseline.

8. The method of claim 1 or 7, wherein The quantifying the statistical deviation of the dynamic statistical baseline from the hydraulic data of the mining excavator further comprises applying an exponentially weighted moving average algorithm to the time series of Mahalanobis distances for further calculation.

9. The method of claim 1, wherein, The determining the abnormal condition of the hydraulic system of the mining excavator based on the statistical deviation comprises: when the statistical deviation exceeds a threshold value preset for the current working condition, determining that the hydraulic system is in an abnormal state.

10. A mining excavator operation abnormality detection system for implementing a mining excavator operation abnormality detection method according to claims 1 to 9, characterized by, The system comprises a data acquisition module, a working condition determination module, a dynamic baseline management module, an abnormality detection module, and an early warning module. The data acquisition module is configured to collect analog or digital signals from the operation data sensor and the hydraulic data sensor, and transmit the data to subsequent modules after preprocessing. The working condition determination module is configured to determine the current working condition of the mining excavator according to the operation data collected by the data acquisition module. The dynamic baseline management module is configured to select a dynamic statistical baseline according to the current working condition of the mining excavator. The abnormality detection module is configured to determine the abnormal state of the hydraulic system of the mining excavator according to the statistical deviation of the dynamic statistical baseline from the hydraulic data collected by the data acquisition module. The early warning module is configured to generate and issue a warning when it is determined that the hydraulic system is in an abnormal state.

Images