Bucket tooth abnormality detection method, bucket tooth abnormality detection device, electronic device, and product

Abstract

This application relates to the field of intelligent monitoring technology for construction machinery, and provides a method, device, electronic equipment, and product for detecting bucket tooth anomalies. The method includes: acquiring a temporal acoustic signal collected by an acoustic acquisition module, wherein the acoustic acquisition module is installed on a rigid structure near the bucket side of the excavator's boom or stick; the temporal acoustic signal includes the acoustic signal at the current moment and at least one acoustic signal acquired at a time earlier than the current moment; constructing a corresponding multidimensional feature vector based on the temporal acoustic signal; calculating a difference index between the multidimensional feature vector and a fingerprint database, wherein the difference index reflects the difference between the multidimensional feature vector and the fingerprint database, and the fingerprint database includes standardized feature vectors of samples in a full-tooth state; and performing bucket tooth anomaly detection on the excavator based on the difference index. This application enables accurate and stable online detection of bucket tooth anomalies.

Description

Technical Field

[0001] This application belongs to the field of intelligent monitoring technology for engineering machinery, and in particular relates to a method for detecting abnormal bucket teeth, a device for detecting abnormal bucket teeth, electronic equipment and products. Background Technology

[0002] Excavator bucket teeth, as vulnerable parts that come into direct contact with soil and rock, are susceptible to damage if broken or detached. This not only reduces digging efficiency but can also allow foreign metal objects to enter downstream equipment such as crushers, causing major accidents and downtime losses. Therefore, how to accurately and reliably detect bucket tooth anomalies online is a pressing technical problem that needs to be solved. Summary of the Invention

[0003] This application provides a method, device, electronic equipment, and product for detecting abnormal bucket teeth, which can accurately and stably detect abnormalities in bucket teeth online.

[0004] In a first aspect, embodiments of this application provide a method for detecting abnormal bucket teeth, including: Acquire acoustic vibration timing signals collected by the acoustic vibration acquisition module. The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the boom or stick of the excavator. The acoustic vibration timing signals include the acoustic vibration signal at the current moment and at least one acoustic vibration signal with an acquisition time earlier than the current moment. Based on the acoustic vibration time-series signal, a corresponding multidimensional feature vector is constructed; Calculate the difference index between the multidimensional feature vector and the fingerprint database. The difference index reflects the difference between the multidimensional feature vector and the fingerprint database, which includes standardized feature vectors of samples with full teeth. Based on the aforementioned difference indicators, the excavator is subjected to bucket tooth anomaly detection.

[0005] In this embodiment, by installing an acoustic vibration acquisition module on the rigid structure near the bucket side of the excavator's boom or stick, and acquiring the acoustic vibration time-series signal collected by the module, a corresponding multi-dimensional feature vector can be constructed based on this signal. The difference index between the multi-dimensional feature vector and a fingerprint database is then calculated. Based on this difference index, online detection of bucket tooth anomalies is achieved. This solution does not rely on images of the bucket teeth's appearance. Even under visual failure conditions such as mud covering, darkness, heavy dust, or obstructed cameras, it can still rely on acoustic vibration time-series signals to detect bucket tooth anomalies, thus enabling accurate and stable online detection.

[0006] In some embodiments of the first aspect, constructing a corresponding multidimensional feature vector based on the acoustic vibration time-series signal includes: Perform a short-time Fourier transform on the acoustic vibration time-series signal to obtain the time-frequency distribution; The acoustic vibration time-series signal is subjected to multi-layer wavelet packet decomposition to calculate the energy ratio characteristics of adjacent frequency bands; Based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic-vibration time-series signal, the multidimensional feature vector is constructed.

[0007] In some embodiments of the first aspect, constructing the multidimensional feature vector based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic-vibration time-series signal includes: Based on the aforementioned time-frequency distribution, the average power spectral density at each frequency point is calculated; The average power spectral density of each frequency point is normalized to the energy spectral density to obtain the energy spectral density of each frequency point. A specific frequency band is divided into multiple characteristic frequency bands, and the specific frequency band is a frequency range that differs between the full tooth state and the abnormal tooth state. The multidimensional feature vector is constructed based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators.

[0008] In some embodiments of the first aspect, constructing the multidimensional feature vector based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators includes: For each of the characteristic frequency bands, the energy spectral densities of the corresponding target frequency points are added together to obtain the energy characteristics of the characteristic frequency band, and each target frequency point is a frequency point located within the characteristic frequency band among the frequency points. The frequency index corresponding to the maximum value of the energy spectral density in the characteristic frequency band is determined as the real-time main peak frequency of the characteristic frequency band; Calculate the spectral centroid based on the average power spectral density and the frequency index of each frequency point; Spectral broadening is calculated based on the spectral centroid, the average power spectral density of each frequency point, and the frequency index of each frequency point. The multidimensional feature vector is constructed based on the energy characteristics of multiple characteristic frequency bands, the spectral centroid, the spectral broadening, the energy ratio characteristics of adjacent frequency bands, the real-time main peak frequencies of multiple characteristic frequency bands, and the time-domain statistical indicators of the acoustic vibration time-series signal.

[0009] In some embodiments of the first aspect, prior to performing bucket tooth anomaly detection on the excavator based on the difference index, the method further includes: A key feature frequency band is determined from multiple feature frequency bands, wherein the key feature frequency band is the frequency range in which the difference is greatest between the full tooth state and the abnormal tooth state; Calculate the offset corresponding to the key feature frequency band, where the offset corresponding to the key feature frequency band is the offset between the real-time main peak frequency of the key feature frequency band and the average of the standard main peak frequencies of the key feature frequency band in the fingerprint database. The step of detecting bucket tooth anomalies in the excavator based on the difference index includes: If the difference index is greater than the difference threshold, and the offset corresponding to the key feature frequency band is greater than the offset threshold, then it is determined that the excavator has an abnormal bucket tooth.

[0010] In some embodiments of the first aspect, the step of detecting bucket tooth anomalies in the excavator based on the difference index includes: The difference index is mapped to the acoustic and vibration anomaly confidence level of the bucket teeth; Based on the images of the bucket teeth of the excavator, the confidence level of visual anomalies of the bucket teeth is determined; The acoustic vibration anomaly confidence score and the visual anomaly confidence score are fused to obtain the fused anomaly confidence score. Based on the difference index and the fused anomaly confidence level, the health status level of the excavator's bucket teeth is determined, and when the bucket tooth health status level indicates that the excavator has a bucket tooth abnormality, an alarm strategy corresponding to the bucket tooth health status level is triggered.

[0011] In some embodiments of the first aspect, the step of detecting bucket tooth anomalies in the excavator based on the difference index includes: Based on the aforementioned difference index, the probability of abnormality in the acoustic vibration detection of the bucket teeth is determined; Based on the multidimensional feature vector, the reliability factor of the acoustic vibration detection is determined; Based on the probability of abnormality of the bucket teeth detected by the acoustic vibration detection and the reliability factor of the acoustic vibration detection, the basic probability assignment of the acoustic vibration evidence body is determined. Based on the excavator's bucket tooth image, determine the probability of bucket tooth anomalies detected by visual inspection and the reliability factor of the visual inspection. Based on the probability of abnormal teeth detected by the visual inspection and the reliability factor of the visual inspection, the basic probability assignment of the visual evidence body is determined. Based on the basic probability assignments of the acoustic and vibration evidence and the basic probability assignments of the visual evidence, the excavator is subjected to bucket tooth anomaly detection.

[0012] In some embodiments of the first aspect, the detection of bucket tooth anomalies in the excavator based on the basic probability assignments of the acoustic and vibration evidence and the visual evidence includes: The Dempster combination rule is used to fuse the basic probability assignments of the acoustic and vibrational evidence and the visual evidence to obtain the fused basic probability assignments. Multiply the fused basic probability assignment of the uncertain state in the fused basic probability assignment by a preset value to obtain the target value; The target value is added to the fused basic probability assignment of the abnormal state in the fused basic probability assignment to obtain the final abnormal probability. Alternatively, subtract the target value and the fused basic probability assignment of the normal state from 1 to obtain the final abnormal probability. If the final anomaly probability is greater than the anomaly probability threshold, then it is determined that the excavator has an abnormal bucket tooth.

[0013] In some embodiments of the first aspect, when the number of acquisition channels for the acoustic vibration timing signal is multiple, the calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Determine the attention weights for multiple acquisition channels; Based on the attention weights of the multiple acquisition channels, the multidimensional feature vectors corresponding to the acoustic and vibration time-series signals of the multiple acquisition channels are weighted and fused to obtain the fused feature vector; Calculate the difference index between the fused feature vector and the fingerprint database.

[0014] In some embodiments of the first aspect, acquiring the acoustic vibration timing signal acquired by the acoustic vibration acquisition module includes: Under the target working condition, the acoustic vibration timing signal collected by the acoustic vibration acquisition module is acquired. The target working condition is any one of the following: the moment the bucket touches the ground, the stable cutting stage, and the no-load lifting stage. The calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Calculate the difference index between the multidimensional feature vector and the target standardized feature vector in the fingerprint database, wherein the target standardized feature vector includes the standardized feature vector of the full-tooth state sample under the target working condition.

[0015] In some embodiments of the first aspect, the method for identifying the moment the bucket touches the ground includes: Monitor the rate of change of Z-axis acceleration at the boom end and the rate of change of pressure in the bucket cylinder; If the rate of change of the Z-axis acceleration at the end of the boom is greater than the acceleration rate of change threshold, and the rate of change of the pressure in the bucket cylinder is greater than the pressure rate of change threshold, then it is determined that the moment the bucket touches the ground has been detected.

[0016] In some embodiments of the first aspect, calculating the difference index between the multidimensional feature vector and the fingerprint database includes: When there are multiple sets of standardized feature vectors for the full-tooth state samples, calculate the mean vector and covariance matrix of the standardized feature vectors for the multiple sets of full-tooth state samples. Based on the multidimensional feature vector, the mean vector, and the covariance matrix, the distance between the multidimensional feature vector and the fingerprint database is calculated, and the distance is determined as the difference index.

[0017] Secondly, embodiments of this application provide a bucket tooth abnormality detection device, comprising: The signal acquisition module is used to acquire the acoustic vibration timing signal acquired by the acoustic vibration acquisition module. The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the boom or stick of the excavator. The acoustic vibration timing signal includes the acoustic vibration signal at the current moment and at least one acoustic vibration signal acquired at a time earlier than the current moment. The feature construction module is used to construct a corresponding multidimensional feature vector based on the acoustic vibration time-series signal; The difference calculation module is used to calculate the difference index between the multidimensional feature vector and the fingerprint database. The difference index reflects the difference between the multidimensional feature vector and the fingerprint database, which includes standardized feature vectors of full-tooth state samples. An anomaly detection module is used to detect bucket tooth anomalies in the excavator based on the difference indicators.

[0018] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the electronic device implements the bucket tooth abnormality detection method as described in any of the first aspects above.

[0019] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a computer, implements the bucket tooth abnormality detection method as described in any one of the first aspects above.

[0020] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when run, causes the bucket tooth abnormality detection method as described in any of the first aspects above to be executed.

[0021] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart of a method for detecting abnormal bucket teeth provided in an embodiment of this application; Figure 2 This is another schematic flowchart of the bucket tooth anomaly detection method provided in the embodiments of this application; Figure 3 This is another flowchart illustrating the method for detecting abnormal bucket teeth provided in the embodiments of this application; Figure 4 This is another schematic flowchart of the bucket tooth anomaly detection method provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the bucket tooth abnormality detection device provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0025] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0029] Excavator bucket teeth are vulnerable components that come into direct contact with soil and rock. Their breakage or detachment not only reduces digging efficiency but can also allow foreign metal objects to enter downstream equipment such as crushers, causing major accidents and downtime losses. Therefore, continuous and reliable online monitoring of bucket tooth abnormalities is necessary.

[0030] Current mainstream solutions employ machine vision technology, using cameras to capture images of the bucket and deep learning models to identify the outline and number of bucket teeth. Some solutions incorporate depth cameras and radar to achieve multimodal fusion. These solutions perform reasonably well under good lighting and with clean bucket tooth surfaces, but face serious limitations in harsh actual mining conditions: mud coatings prevent cameras from capturing only irregular "mud-covered" outlines, making it impossible to distinguish the true shape of the bucket teeth; low light, strong backlighting, and high concentrations of dust significantly reduce image quality; lens contamination and frost further deteriorate camera imaging. Even with infrared or high dynamic range (HDR) enhancement, it is difficult to overcome the problem of mud blending with background textures.

[0031] Furthermore, vision-based solutions rely on changes in the external profile and are insensitive to early damage such as internal cracks and loosening of the tooth roots, lacking predictive maintenance capabilities. Solutions that use process quantities such as current and voltage to reflect overall load changes are extremely insensitive to minute structural changes in individual bucket teeth.

[0032] From a physical perspective, the bucket teeth, bucket, boom, and arm constitute a mechanical vibration system. When the mass and stiffness of the bucket teeth change, the natural frequency, damping characteristics, and impact response spectrum of the end effector undergo measurable changes. These changes propagate to the rigid components in the form of elastic waves and can be acquired by an acoustic vibration acquisition module. Current technology has not fully utilized this mechanical vibration characteristic and lacks a system solution for jointly modeling acoustic vibration time-series signals into a multi-dimensional feature vector and combining it with adaptive triggering based on operating conditions for anomaly detection.

[0033] Based on this, the embodiments of this application acquire acoustic and vibration timing signals by directly installing an acoustic and vibration acquisition module on the boom or stick, effectively avoiding the limitations of visual obstruction and harsh environments. By constructing a multi-dimensional feature vector integrating time and frequency domain features and calculating the difference index between it and a fingerprint database based on a large number of normal samples (i.e., full-tooth state samples), sensitive capture and quantitative evaluation of minute and essential changes in the state of the bucket teeth are achieved. This method can achieve stable and accurate non-visual online anomaly detection under various actual operating conditions of excavators, thereby providing timely warnings of the risk of bucket tooth breakage or detachment, and preventing the resulting efficiency reduction and secondary equipment damage accidents.

[0034] The bucket tooth anomaly detection method provided in this application embodiment can be applied to electronic devices such as servers, tablet computers, excavator controllers or additional embedded processing units, desktop computers, laptop computers, ultra-mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs). This application embodiment does not impose any restrictions on the specific type of electronic device.

[0035] To illustrate the technical solution of this application, specific embodiments are described below.

[0036] Please see Figure 1 , Figure 1 The diagram illustrates a flowchart of a bucket tooth anomaly detection method provided in an embodiment of this application. This is an example, not a limitation, and the method is applied to the controller of an excavator. The method includes the following steps: Step 101: Obtain the acoustic vibration timing signal acquired by the acoustic vibration acquisition module.

[0037] The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the excavator's boom or stick. The acoustic vibration timing signal includes the acoustic vibration signal at the current moment and at least one acoustic vibration signal acquired at a time earlier than the current moment. The acoustic vibration signal can include vibration signals (e.g., triaxial vibration signals) and acoustic emission signals.

[0038] In some embodiments, the acoustic vibration acquisition module can acquire vibration signals at a recommended sampling rate of 20 kHz according to a working condition trigger command. When higher frequency resolution is required, the sampling rate can be configured in the range of 20 kHz to 50 kHz. At the moment of impact, the acoustic emission signal is acquired briefly at a sampling rate of 0.5 MHz to 1 MHz. The vibration signal and the acoustic emission signal are the structural response at the installation location of the acoustic vibration acquisition module.

[0039] In some embodiments, the acoustic vibration acquisition module can perform anti-aliasing filtering, pre-trigger buffering, and timestamp marking on the acquired vibration signals and acoustic emission signals, and maintain millisecond-level time alignment with the excavator control system through synchronization pulses or time synchronization protocols.

[0040] In this embodiment, the vibration signal can be excited by a constant current source and then enter the signal conditioning circuit. It is then passed through a high-pass filter to remove DC bias and low-frequency drift, and then through a low-pass filter for anti-aliasing processing. The typical cutoff frequency of the high-pass filter is 1Hz, and the cutoff frequency of the low-pass filter is preferably set at approximately 40% to 50% of the sampling frequency. At a recommended sampling frequency of 20 kHz, the low-pass cutoff frequency is approximately 8 kHz to 10 kHz, which is higher than the upper limit of the analysis frequency of this scheme by 5 kHz, satisfying the Nyquist sampling and anti-aliasing requirements. The conditioned analog signal is processed by a 24-bit Sigma-Aldrich signal. The Delta-type ADC converts signals to digital signals at a recommended sampling frequency of 20 kHz. For higher resolution or wider bandwidth requirements, the sampling rate can be increased to 50 kHz, depending on hardware capabilities. The acoustic emission signal uses an independent high-speed acquisition channel, allowing for short-term acquisition of several sampling points at a sampling rate of 0.5 MHz to 1 MHz only when an impact event is detected, for high-frequency acoustic emission characteristic analysis.

[0041] The pre-triggered buffer refers to storing a short segment of the most recent acoustic vibration signal in advance. The acoustic vibration acquisition module can place the acoustic vibration signal from a recent period in a circular buffer. Once a ground impact is detected or the trigger condition is met, it not only saves the acoustic vibration signal after the trigger but also the acoustic vibration signal in the buffer before the trigger. This allows for the capture of the transition process before the event occurs, facilitating subsequent complete feature extraction and alignment analysis.

[0042] In some embodiments, the controller can maintain a circular buffer in memory to continuously pre-acquire acoustic and vibration signals. When the trigger condition is met, a certain length of historical data in the buffer, along with new data after the trigger, is used as an input for subsequent processing within a time window. The typical time window length is 0.2 seconds to 2 seconds, corresponding to a number of sampling points. ,in Sampling frequency, The time window is the length of the time window. The working condition time window can refer to the total duration set to capture the entire digging action (such as a complete bucket impact).

[0043] In some embodiments, the acoustic vibration acquisition module may include several vibration sensors and acoustic emission sensors. Vibration sensors are used to acquire vibration signals. Acoustic emission sensors are used to acquire acoustic emission signals. Vibration sensors can be mounted on a rigid structure near the boom-stalk connection pin, a location that effectively receives vibration energy from the bucket teeth while being relatively far from mud erosion in the bucket tooth area, balancing signal-to-noise ratio and reliability. Acoustic emission sensors can be mounted in the welded area near the root of the bucket teeth or the tooth seat, sensitive to high-frequency stress waves from 30 kHz to 300 kHz, and can be used to capture transient elastic waves of crack propagation within the bucket teeth, enabling early damage warning.

[0044] Vibration sensors and acoustic emission sensors can be fixed with polished metal mounting surfaces and magnetic or threaded bases, and the protection level can be improved by using stainless steel protective covers and silicone sealing rings. The target protection level can reach IP67 to adapt to the harsh environment of mines.

[0045] This application does not limit the number of vibration sensors and acoustic emission sensors.

[0046] This application does not limit the specific type of vibration sensor. For example, the vibration sensor is an integrated electronic piezoelectric (IEPE) type high-frequency accelerometer (e.g., an IEPE type single-axis or triaxial piezoelectric accelerometer with a typical range of ±500 g).

[0047] This application does not limit the specific type of acoustic emission sensor. For example, the acoustic emission sensor may be a broadband acoustic emission sensor.

[0048] In some embodiments, the frequency response range of the vibration sensor can be from 0.5 Hz to 10 kHz.

[0049] In this embodiment, the frequency response range of the vibration sensor is designed to cover the main energy components of the excavator end-effector vibration. The lower limit is extended to 0.5 Hz to include possible slow motion components; the upper limit is set at approximately 10 kHz, primarily based on two considerations: firstly, the vibration energy generated by structural stress changes and resonance during excavation operations is mainly distributed in the low to mid-frequency range (typically within several kilohertz); secondly, transient impacts such as bucket teeth contacting the ground and cutting collisions can excite high-frequency modal responses in the structure. Setting the upper limit at 10 kHz effectively includes these high-frequency components that may contain state information within the acquisition range and provides sufficient transition band for anti-aliasing filtering in subsequent signal conditioning circuits, avoiding the erroneous filtering out of effective high-frequency information.

[0050] In some embodiments, the operating frequency band of the acoustic emission sensor can be from 30 kHz to 300 kHz.

[0051] In this embodiment, the operating frequency band of the acoustic emission sensor is a typical effective operating range for acoustic emission sensors used in engineering for monitoring damage to metal structures. This selection is based on a trade-off between the physical characteristics of acoustic emission signals and engineering practice: First, the elastic waves (acoustic emission signals) released by damage mechanisms such as the initiation and propagation of microcracks inside the bucket teeth and tooth surface peeling have their energy mainly concentrated in this frequency band, making them easy for the sensor to capture. Second, this frequency band can effectively avoid the strong low-frequency vibration noise of the excavator host and the noise of the hydraulic system, thereby obtaining a high signal-to-noise ratio. If the frequency band is too low, the signal is easily drowned out by background mechanical noise; if the frequency band is too high, the propagation attenuation of high-frequency elastic waves in the steel structure increases sharply, placing extremely stringent requirements on the sensor installation coupling quality and the performance of the acquisition link, resulting in low cost-effectiveness. Therefore, 30 kHz to 300 kHz is a preferred frequency band that strikes a balance between monitoring sensitivity, signal propagation efficiency, and system implementation cost.

[0052] Step 102: Construct the corresponding multidimensional feature vector based on the acoustic vibration time sequence signal.

[0053] Step 103: Calculate the difference index between the multidimensional feature vector and the fingerprint database.

[0054] The difference index reflects the degree of difference or deviation between the multidimensional feature vector and the fingerprint database. The fingerprint database includes standardized feature vectors of samples with full teeth.

[0055] Step 104: Based on the difference index, perform bucket tooth anomaly detection on the excavator.

[0056] In some embodiments, the difference index can be compared with a preset threshold. If the difference index is greater than the preset threshold, it is determined that the excavator has a bucket tooth abnormality; if the difference index is less than or equal to the preset threshold, it is determined that the excavator does not have a bucket tooth abnormality. Optionally, the preset threshold can be set according to actual needs or empirical values.

[0057] In this embodiment, based on the difference index, the excavator is subjected to bucket tooth anomaly detection. Without relying on the appearance image of the bucket teeth, even under visual failure conditions such as mud covering, no light at night, heavy dust, and camera obstruction, the detection of missing and severely worn bucket teeth can still be completed solely by the acoustic vibration channel. Under typical mining conditions, the recall rate for missing teeth identification can be maintained at around 95%, reducing the missed detection rate in visual blind spots from nearly 100% to a significantly acceptable level.

[0058] In this embodiment, by installing an acoustic vibration acquisition module on the rigid structure near the bucket side of the excavator's boom or stick, and acquiring the acoustic vibration time-series signal collected by the module, a corresponding multi-dimensional feature vector can be constructed based on this signal. The difference index between the multi-dimensional feature vector and a fingerprint database is then calculated. Based on this difference index, online detection of bucket tooth anomalies is achieved. This solution does not rely on images of the bucket teeth's appearance. Even under visual failure conditions such as mud covering, darkness, heavy dust, or obstructed cameras, it can still rely on acoustic vibration time-series signals to detect bucket tooth anomalies, thus enabling accurate and stable online detection.

[0059] In some embodiments of this application, acquiring the acoustic vibration timing signal acquired by the acoustic vibration acquisition module includes: Under the target working condition, the acoustic and vibration timing signal collected by the acoustic and vibration acquisition module is acquired. The target working condition is any one of the following: the moment the bucket touches the ground, the stable cutting stage, and the no-load lifting stage. Calculate the difference metrics between the multidimensional feature vector and the fingerprint database, including: Calculate the difference index between the multidimensional feature vector and the target standardized feature vector in the fingerprint database. The target standardized feature vector includes the standardized feature vector of the full-tooth state sample under the target working condition.

[0060] At the moment the bucket touches the ground, it changes from a free-moving state to contact with the ground, generating a broadband impact signal that can excite the high-frequency modal response of the structure.

[0061] During the stable cutting phase, the bucket teeth are in continuous contact with the material, and the system is in a quasi-steady state of forced vibration. This is beneficial for observing the energy distribution and spectral centroid changes in the low and mid-frequency bands. This condition can be identified by detecting that the bucket cylinder pressure remains within the working pressure range (e.g., 15MPa to 25MPa) for several seconds, combined with the amplitude of attitude angle changes.

[0062] During the unloaded lifting phase, the bucket has left the material, and the robotic arm gradually returns to an unloaded state. The structure exhibits free vibration decay characteristics, making it more suitable for observing changes in natural frequency and damping. This condition can be identified by monitoring the extension of the boom cylinder and the bucket cylinder pressure falling below a low-pressure threshold (e.g., below 5 MPa).

[0063] In this embodiment, the excavator's controller can analyze sensor data such as boom angle, stick angle, boom and bucket cylinder pressure, and inertial measurement unit (IMU) attitude to identify typical working conditions such as ground impact, stable cutting, and unloaded lifting. High-frequency acoustic and vibration acquisition is triggered only within these "effective excitation windows," avoiding the acquisition of large amounts of invalid noise during unloaded slewing and traveling phases. This ensures that the multi-dimensional feature vectors parsed from the acoustic and vibration time-series signals can accurately and sensitively reflect changes in the mass and stiffness of the bucket teeth, rather than being overwhelmed by random noise during operation. This significantly enhances the sensitivity and accuracy of the anomaly detection algorithm for missing, worn, and early-age damage to bucket teeth, while effectively suppressing false alarms caused by signal quality fluctuations. Ultimately, this ensures that the entire system can still output stable and reliable anomaly detection results even under harsh working conditions where visual failure occurs. The boom angle can refer to the boom's reference direction relative to the upper body / frame (equivalent to the boom's attitude angle relative to the machine body around its hinge point). The stick angle can refer to the angle between the stick and the boom (equivalent to the attitude angle of the stick about its hinge point relative to the boom).

[0064] In this embodiment, by acquiring acoustic and vibration timing signals under any of the following working conditions—the instant the bucket touches the ground, the stable cutting stage, and the unloaded lifting stage—a multi-dimensional feature vector with discriminative power under different working conditions can be constructed. After constructing the multi-dimensional feature vector under the target working condition, the difference index between this multi-dimensional feature vector and the standardized feature vector of the full-tooth state sample under the target working condition can be calculated to accurately detect bucket tooth anomalies under the target working condition.

[0065] In some embodiments of this application, the method for identifying the moment the bucket touches the ground includes: Monitor the rate of change of Z-axis acceleration at the boom end and the rate of change of pressure in the bucket cylinder; If the rate of change of acceleration at the Z-axis of the boom end is greater than the acceleration rate of change threshold, and the rate of change of pressure in the bucket cylinder is greater than the pressure rate of change threshold, then the moment the bucket touches the ground is detected.

[0066] In this embodiment, the trigger flag for the moment the bucket touches the ground can be determined based on the rate of change of the Z-axis acceleration at the end of the boom and the rate of change of the pressure in the bucket cylinder. If the trigger flag is 1, it is determined that the moment the bucket touches the ground has been detected.

[0067] The formula for calculating the trigger signal at the moment the bucket touches the ground is as follows:

[0068] in, Z-axis acceleration at the end of the boom (m / s²); Hydraulic pressure of the bucket cylinder (MPa); The threshold for the rate of change of acceleration; This is the threshold for the rate of change of pressure.

[0069] Optionally, the acceleration rate of change threshold and the pressure rate of change threshold can be set according to actual needs or empirical values. For example, a typical value for the acceleration rate of change threshold is about 1.5 g / s, and a typical value for the pressure rate of change threshold is about 50 MPa / s.

[0070] In some embodiments of this application, such as Figure 2 As shown, constructing a corresponding multidimensional feature vector based on the acoustic vibration time-series signal can include steps 201 to 203.

[0071] Step 201: Perform a short-time Fourier transform on the acoustic vibration time sequence signal to obtain the time frequency distribution.

[0072] For single-channel acoustic vibration timing signals The time-frequency distribution is calculated using the Short-Time Fourier Transform (STFT). Let the analysis window length be... Window moved to , No. Frame, First The STFT at each frequency point is:

[0073] in, For the first Frame in frequency index Complex spectral values ​​at; For time frame indexing; Frequency index; For length is Windowing functions (such as Hanning windows or flat-top windows); The analysis window length (number of sampling points, typically 512 to 2048); For frame shift, satisfying , The overlap length is used for short-time Fourier transform (SFT) analysis. The analysis window length is the length of a small slice of signal within the operating time window used for spectral calculation. Frame shift refers to how much the analysis window slides forward each time during SFT analysis.

[0074] Step 202: Perform multi-level wavelet packet decomposition on the acoustic vibration time sequence signal and calculate the energy ratio characteristics of adjacent frequency bands.

[0075] To obtain finer-grained multi-scale frequency band features, multi-level wavelet packet decomposition was performed on the preprocessed acoustic-vibration time-series signal. The db6 wavelet basis was used for decomposition. Layer decomposition (e.g.) ), dividing the frequency band into One frequency band. Layer The wavelet packet coefficients corresponding to each frequency band are denoted as follows: , No. Layer Band energy of each frequency band It can be represented as:

[0076] To mitigate the impact of absolute amplitude variations, the energy ratio characteristics of adjacent frequency bands are constructed:

[0077] in, For the first The frequency band and the first The energy ratio of each frequency band; To prevent division by zero of small constants (e.g.) ).

[0078] Step 203: Construct a multidimensional feature vector based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic-vibration time-series signal.

[0079] The time-domain statistical indicators of acoustic vibration time-series signals include, but are not limited to, the root mean square (RMS) of the acoustic vibration time-series signal (reflecting the overall intensity or energy level of this segment of the acoustic vibration time-series signal, biased towards describing the overall vibration strength), peak factor, kurtosis (reflecting whether there are relatively sharp impact components or abnormal pulses in this segment of the acoustic vibration time-series signal, biased towards describing whether a sudden impact has occurred), skewness, etc.

[0080] In this embodiment, the short-time Fourier transform provides the time-frequency localization characteristics of the signal, effectively capturing frequency component changes in transient events such as tooth impact and cutting. Wavelet packet decomposition and adjacent frequency band energy ratio characterize the distribution pattern of frequency band energy at a finer scale, being extremely sensitive to minute changes in parameters such as structural stiffness and mass, and weakening the impact of overall signal amplitude fluctuations through ratio calculation. Time-domain statistical indicators directly reflect the signal strength and impact characteristics. These three complement each other, ensuring that regardless of whether the tooth anomaly manifests as frequency shift, energy redistribution, or changes in time-domain impact characteristics, it can be significantly reflected in the feature vector. This provides rich and robust input for subsequent accurate comparison with the fingerprint database, fundamentally improving the sensitivity, accuracy, and system adaptability of anomaly detection.

[0081] In some embodiments of this application, such as Figure 3 As shown, a multidimensional feature vector is constructed based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of acoustic vibration time-series signals. This may include steps 301 to 304.

[0082] Step 301: Calculate the average power spectral density at each frequency point based on the time-frequency distribution.

[0083] The controller can calculate the average power spectral density at each frequency point based on the STFT results:

[0084] in, For the number of time frames, Frequency Index The average power spectral density at the corresponding frequency point.

[0085] Step 302: Normalize the average power spectral density of each frequency point to the energy spectral density to obtain the energy spectral density of each frequency point.

[0086] To eliminate the impact of differences in overall energy levels under different operating conditions, Normalized to energy spectral density:

[0087] in, Frequency Index The energy spectral density at the corresponding frequency point.

[0088] Step 303: Divide the specific frequency band into multiple characteristic frequency bands. The specific frequency band is the frequency range that differs between the full tooth state and the abnormal tooth state.

[0089] In some embodiments, a specific frequency band can be determined by combining mechanical structural characteristics and measured data. Specifically, acoustic and vibration timing signals of normal and abnormal bucket teeth under typical working conditions can be collected over a wide frequency range, such as ground impact, stable cutting, and no-load lifting. Then, the response differences in different frequency bands are compared. If certain frequency bands show significant changes in the full-tooth state and the abnormal bucket teeth state, and these changes are relatively stable in multiple samplings and less affected by environmental noise, then these frequency bands can be considered as the focus of subsequent analysis. If certain frequency bands mainly reflect changes in the overall machine attitude, hydraulic fluctuations, engine noise, or have exceeded the effective response range of the sensors, then these frequency bands are generally not considered as the focus of analysis. The frequency range of a specific frequency band is related to the bucket teeth. Because missing, worn, or stiffness-changed bucket teeth will change the vibration characteristics of the bucket end structure, causing changes in the energy distribution, peak position, or impact response of certain frequency bands. However, it is not only determined by the bucket teeth alone, but is also affected by factors such as the bucket and boom structure, sensor installation location, working condition type, sampling rate, and on-site noise environment. This frequency range is a valid analysis range that is related to changes in the bucket teeth's condition and has been selected through actual testing. That is, the frequency range of a specific band is determined through broadband acquisition and comparative experiments, retaining those frequency bands that are most sensitive to bucket teeth anomalies, have good repeatability, and have low noise interference. This range is related to the bucket teeth's condition and is also affected by the overall machine structure, sensor arrangement, and operating conditions.

[0090] Step 304: Construct a multidimensional feature vector based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and time-domain statistical indicators.

[0091] In this embodiment, specific frequency bands exhibiting differences between the full-tooth state and the abnormal-tooth state are precisely selected from the time-frequency distribution obtained by short-time Fourier transform as feature frequency bands, and their energy is calculated. This ensures that the extracted frequency domain features can directly and sensitively capture the core frequency response variations caused by changes in the quality and stiffness of the bucket teeth, greatly enhancing the specificity and state distinguishability of the features. Furthermore, the average power spectral density characterizing the absolute energy level, the normalized energy spectral density eliminating the influence of overall energy, the wavelet packet adjacent frequency band energy ratio revealing the fine-scale energy distribution pattern, and the time-domain statistical indicators reflecting signal strength and impact characteristics are organically integrated to construct a comprehensive and complementary multi-dimensional feature vector. This vector can comprehensively and stably characterize the dynamic state of the bucket tooth-structure system from multiple levels, including absolute energy value, relative distribution, multi-scale details, and time-domain waveforms. It provides robust and information-rich input for subsequent high-precision comparison with the fingerprint database, and is a key feature guarantee for this solution to achieve accurate and reliable anomaly detection under visual failure conditions.

[0092] In some embodiments of this application, a multidimensional feature vector is constructed based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and time-domain statistical indicators, including: For each characteristic frequency band, the energy spectral density of each corresponding target frequency point is added together to obtain the energy characteristic of the characteristic frequency band. Each target frequency point is the frequency point located within the characteristic frequency band among all frequency points. The frequency index corresponding to the maximum value of the energy spectral density in the characteristic frequency band is determined as the real-time main peak frequency of the characteristic frequency band; Calculate the spectral centroid based on the average power spectral density and frequency index of each frequency point; Spectral broadening is calculated based on the spectral centroid, the average power spectral density at each frequency point, and the frequency index at each frequency point. A multidimensional feature vector is constructed based on the energy characteristics, spectral centroid, spectral broadening, energy ratio characteristics of adjacent frequency bands, real-time main peak frequencies of multiple characteristic frequency bands, and time-domain statistical indicators of acoustic vibration time-series signals.

[0093] The controller can select a specific frequency band from 0 to 10 on the frequency axis. (For example, 5kHz) is divided into several characteristic frequency bands, the first... The frequency index range of each characteristic frequency band is Its normalized energy is:

[0094] in, For the first Energy characteristics of each characteristic frequency band (dimensionless, value range from 0 to 1); , For the first The minimum and maximum frequency indices within each characteristic frequency band.

[0095] Within a specific frequency band, the structural modal resonance peaks are highly sensitive to changes in the mass and stiffness of the bucket teeth. This is based on the normalized energy spectral density. Find the frequency index corresponding to the maximum value within each selected characteristic frequency band. This is the main peak frequency of this frequency band. For the full-tooth state, statistics can be used. The mean and variance; during operation, when When a continuous shift occurs, it indicates that the corresponding mode of that frequency band has changed, which can be used as an auxiliary anomaly criterion.

[0096] To characterize the centroid and dispersion of the overall energy distribution of the spectrum, the spectral centroid is defined. as follows:

[0097] Define spectrum broadening as follows:

[0098] When the bucket teeth fall off or become severely thinned, the end mass decreases, the equivalent stiffness / mass ratio increases, and the system's natural frequency shifts to higher frequencies overall, manifesting as a spectral centroid. Shifting to higher frequencies, while structural symmetry is disrupted, resulting in spectral broadening. It will generally increase in size.

[0099] Based on the energy characteristics, spectral centroid, spectral broadening, energy ratio characteristics of adjacent frequency bands, real-time main peak frequency of multiple characteristic frequency bands, and time-domain statistical indicators of acoustic vibration time-series signals, a multidimensional feature vector is constructed. This can refer to concatenating at least two of the following: energy characteristics, spectral centroid, spectral broadening, energy ratio characteristics of adjacent frequency bands, real-time main peak frequency, and time-domain statistical indicators to obtain a multidimensional feature vector.

[0100] As an example, and not a limitation, the operating condition category is constructed by combining the energy characteristics, spectral centroid, spectral broadening, energy ratio characteristics of adjacent frequency bands, real-time main peak frequencies of multiple characteristic frequency bands, and time-domain statistical indicators of acoustic and vibration time-series signals. The multidimensional feature vectors for a certain time window are as follows:

[0101] in, For working conditions The eigenvectors below; The number of selected characteristic frequency bands; is the number of dimensions of the energy ratio feature; Root mean square in time-domain statistical indicators; Kurtosis is a time-domain statistical index. The root mean square (RMS) and kurtosis in time-domain statistical indices can supplement the characterization of changes in the state of the tusk from the perspectives of overall energy and anomalous impact, respectively.

[0102] In this embodiment, when detecting abnormalities in bucket teeth, the acoustic and vibration frequency domain fingerprint is used. Instead of relying on the appearance image of the bucket teeth, the frequency response characteristics under typical working conditions such as excavator ground contact, cutting, and no-load are utilized to construct a working condition frequency domain fingerprint that can characterize the changes in the quality and stiffness of the bucket teeth through features such as multi-scale frequency band energy, spectral centroid and broadening, energy ratio, and modal peak frequency. This avoids the problems of visual occlusion and illumination changes from a physical mechanism perspective.

[0103] It should be understood that the construction method of the standardized feature vector of the full-tooth state sample is the same as the construction method of the multi-dimensional feature vector mentioned above.

[0104] In some embodiments of this application, before performing bucket tooth anomaly detection on the excavator based on difference indicators, the following steps are also included: Key feature frequency bands are determined from multiple feature frequency bands. The key feature frequency band is the frequency range with the greatest difference between the full tooth state and the abnormal tooth state. Calculate the offset corresponding to the key feature frequency band. The offset corresponding to the key feature frequency band is the offset between the real-time main peak frequency of the key feature frequency band and the average of the standard main peak frequencies of the key feature frequency bands in the fingerprint database. Based on the difference indicators, abnormal detection of bucket teeth is performed on excavators, including: If the difference index is greater than the difference threshold, and the offset corresponding to the key feature frequency band is greater than the offset threshold, then it is determined that the excavator has abnormal bucket teeth. If the difference index is less than or equal to the difference threshold, or the offset corresponding to the key feature frequency band is less than or equal to the offset threshold, then it is determined that the excavator does not have any bucket tooth abnormalities.

[0105] The aforementioned key characteristic frequency bands can refer to those frequency bands that are most sensitive to changes in the state of the bucket teeth, have good repeatability, and are less affected by background noise, selected from multiple characteristic frequency bands. For example, on a certain model, certain local frequency bands in the range of several hundred hertz to several thousand hertz may be found to be more sensitive to missing, loose, or stiffness changes in bucket teeth, and these local frequency bands can be used as key characteristic frequency bands.

[0106] To enhance sensitivity to specific modal variations, an offset can be defined for the main peak frequency of certain key characteristic bands:

[0107] in, For real-time main peak frequency, This represents the average peak frequency of this frequency band in a full-tooth fingerprint (i.e., the average standard peak frequency of the key feature band). When Approaching the threshold A larger value can increase the confidence level of the conclusion about missing teeth; conversely, a smaller value may indicate a transient interference.

[0108] In this embodiment, anomaly detection is performed by multi-dimensional feature vectors and modal peak frequency shifts. Combined with high-frequency acoustic emission characteristics, a warning of suspicious anomalies can be given in the early stage when the bucket teeth have not completely broken but the stiffness has decreased significantly or internal cracks have expanded. Compared with the solution that relies solely on vision, potential risks can be detected one shift or even several days in advance, reducing sudden tooth breakage accidents.

[0109] In some embodiments of this application, when there are multiple acquisition channels for acoustic vibration timing signals, the calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Determine the attention weights for multiple acquisition channels; Based on the attention weights of multiple acquisition channels, the multi-dimensional feature vectors corresponding to the acoustic and vibration time-series signals of multiple acquisition channels are weighted and fused to obtain the fused feature vector; Calculate the difference index between the fused feature vector and the fingerprint database.

[0110] In this embodiment, a channel attention mechanism can be used to adaptively weight and fuse the multidimensional feature vectors of multiple acquisition channels to adapt to differences in sensor installation locations and signal quality.

[0111] When multiple acceleration sensors and acoustic emission sensors are installed on the boom, stick, and other components, multi-channel information can be used to improve sensitivity to unilateral tooth loss, asymmetrical wear, and localized cracks. The controller calculates the feature vector of each channel within each operating time window. The fused feature vector is obtained through a channel attention mechanism. :

[0112] in, This refers to the number of data acquisition channels. For the first The attention weights of each acquisition channel satisfy the following conditions: .

[0113] The controller can automatically adjust the attention weights based on the signal-to-noise ratio (SNR) of each acquisition channel, its historical stability, and the deviation of the current feature vector from the mean in the fingerprint database. This reduces the negative impact of channels with poor installation conditions or unstable signal quality on the overall judgment results. The historical stability of each acquisition channel depends primarily on the stability of its signal quality and feature performance over a period of time. For example, whether the noise level is consistently high, whether the amplitude frequently drifts, whether it is prone to saturation or loss, and whether the features extracted under similar operating conditions fluctuate excessively. For instance, if an acquisition channel has a consistently high SNR and the main peak position and energy distribution extracted under multiple ground contact or cutting conditions are relatively stable, then the attention weight of this acquisition channel can be relatively high. Conversely, if another acquisition channel experiences high noise and significant feature fluctuations due to loose installation, poor coupling, or mud interference, its attention weight will be correspondingly reduced. Furthermore, if only one acquisition channel suddenly deviates significantly from the normal baseline while other acquisition channels remain relatively normal, the influence of that acquisition channel will generally be reduced. If multiple stable acquisition channels simultaneously exhibit similar offsets, it is more likely to indicate a genuine structural change and will not be simply treated as a bad channel.

[0114] For acoustic emission channels, high-frequency acoustic emission signals can be monitored during impact events or cutting processes to extract energy, event counts, and arrival time characteristics within a specific frequency band. Crack propagation inside bucket teeth generates short-duration high-frequency stress waves, whose characteristic frequency band differs from ordinary contact noise. The controller can set a crack warning threshold based on the statistical distribution of acoustic emission characteristics. When the acoustic emission event count and signal energy continuously and abnormally increase over a period of time, even if the Mahalanobis distance has not exceeded the missing tooth threshold, an "early damage warning" can be issued, allowing for advance inspection and replacement of bucket teeth. Acoustic emission events can refer to short-duration high-frequency acoustic emission events occurring during ground impact, hard material impact, stable cutting friction, and local crack propagation. The acoustic emission event count is the number of such short-duration high-frequency acoustic emission events identified within an analysis time window. Signal energy refers to the high-frequency signal energy collected by the acoustic emission channel within the target frequency band. Its calculation method involves first limiting the acoustic emission signal to the target frequency band, and then accumulating or integrating the square of the signal amplitude over the duration of a single event or the entire operating time window to obtain the energy value of that event or time window. In practice, it can both count the energy of a single acoustic emission event and accumulate or average the energy of multiple acoustic emission events within a working time window to determine whether there is a continuous abnormal increase.

[0115] In this embodiment, under a multi-sensor configuration, the channel attention mechanism is used to adaptively weight and fuse features from different installation locations and different types of sensors (accelerometer and acoustic emission sensor), which improves the sensitivity to unilateral missing teeth, asymmetric wear and local cracks, while suppressing the disturbance of unstable signal quality channels to the overall decision.

[0116] In some embodiments, during the fingerprint database initialization phase, the mean value of the sample features in the full-tooth state is calculated dimension by dimension. and standard deviation Furthermore, zero-mean unit variance standardization was used to normalize each dimension of the features in order to reduce the impact of differences in different dimensions and magnitudes.

[0117] In some embodiments of this application, such as Figure 4 As shown, calculating the difference index between the multidimensional feature vector and the fingerprint database may include steps 401 and 402.

[0118] Step 401: When there are multiple sets of standardized feature vectors for samples in the full-tooth state, calculate the mean vector and covariance matrix of the standardized feature vectors for multiple sets of samples in the full-tooth state.

[0119] For operating condition categories ,collect Standardized feature vector of samples in full-tooth state Calculate its mean vector Covariance Matrix .

[0120]

[0121]

[0122] Step 402: Based on the multidimensional feature vector, mean vector and covariance matrix, calculate the distance between the multidimensional feature vector and the fingerprint database, and determine the distance as the difference index.

[0123] This application does not limit the type of the aforementioned distance. For example, the aforementioned distance can be Mahalanobis distance, Euclidean distance, cosine distance, etc.

[0124] Since the feature vectors in this application are multidimensional, with different dimensions and often correlated with each other, Mahalanobis distance can take these factors into account when comparing the differences between real-time and standardized feature vectors, making it more suitable for the current scenario.

[0125] During the operation phase, for the current working conditions The observed multidimensional feature vector Define its Mahalanobis distance relative to the full-tooth fingerprint distribution. :

[0126] This indicates the degree of outlierness of the observed multidimensional feature vector in the fingerprint distribution.

[0127] In some embodiments, two thresholds can be set based on the statistical distribution of the full-tooth state samples. and ,when When it is judged as normal, when When it is judged as suspicious or early abnormal, when If it is determined to be a serious abnormality, it should be treated as a risk of tooth loss.

[0128] In the unsupervised fingerprint database scheme of this embodiment, the standardized feature vector set under the full tooth state can be statistically modeled according to the working condition category to construct the mean vector and covariance matrix. The Mahalanobis distance is used to measure the deviation of the real-time feature vector from the fingerprint database, so as to realize anomaly detection that does not require a large number of missing tooth samples.

[0129] In this embodiment, fingerprint database modeling categorized by working conditions and unsupervised detection using Mahalanobis distance significantly reduce the reliance on labeling a large number of abnormal samples. In the initial stage of system deployment, only a limited number of full-tooth normal working condition data need to be collected to complete the calibration, shortening the debugging cycle from installation to stable operation, and providing a mechanism for subsequent online updates to adapt to different mining environments and individual equipment differences.

[0130] In some embodiments, in a supervised approach, the normalized multidimensional feature vector can be fed into a lightweight one-dimensional convolutional neural network. After several convolutional and pooling layers, high-level features are extracted, and then the bucket tooth state category and anomaly probability are output through a fully connected layer and Softmax. To address the class imbalance problem caused by scarce anomalous samples, a weighted cross-entropy loss function can be used, and an online fine-tuning mechanism with experience replay can be designed to allow the model to gradually adapt to different machine types and operating conditions. This supervised approach can serve as a supplementary implementation for unsupervised fingerprint distance determination, improving the recognition capability of complex patterns when data volume is sufficient.

[0131] In some embodiments, to accurately align the acoustic and vibration signals with the operating conditions and bucket tooth images, the internal clock can be periodically calibrated using synchronization pulses or time synchronization protocols. Let the control system timestamp be... The local timestamp of the sound and vibration acquisition module is A simple linear correction model can be used:

[0132] in, For system timestamps (ms); This is the local timestamp (ms) for the acoustic vibration acquisition module. This is the clock ratio correction factor (close to 1). The time offset (ms, typically in the range of ) 5 to 5).

[0133] The controller can estimate through periodic synchronization messages. , This will control the time alignment error between the acoustic and vibration signals and the operating conditions to within 2 ms, providing a foundation for subsequent database construction based on operating conditions and multimodal synchronization.

[0134] The aforementioned periodic synchronization messages refer to time synchronization messages specifically sent by the controller to the acoustic vibration acquisition module. The controller periodically sends these messages. Each time the acoustic vibration acquisition module receives a time synchronization message, it records two things: the time given in the message (the controller's current time) and its own local time when it receives the message. By accumulating multiple sets of corresponding times, the relationship between the two clocks can be determined. If, over a period of time, the local time of the acoustic vibration acquisition module increases slightly faster or slower than the controller's current time, it indicates a difference in the clock's running speed. If the two clocks consistently differ by a small amount, it indicates a fixed time deviation. Statistical estimations can be made based on multiple sets of corresponding points over a recent period, and these two types of errors can be continuously updated to convert the local time recorded by the acoustic vibration acquisition module to the controller's unified time reference. Considering the potential for fixed delays and minor jitter during communication, methods such as sliding windows, averaging filtering, or outlier removal are used to make the estimation results more stable. After this processing, the timestamp of the acoustic and vibration signal can be better aligned with the working condition and the bucket tooth image, which meets the needs of subsequent working condition triggering, correlation analysis and multi-source fusion.

[0135] In embodiments equipped with a visual inspection unit, the acoustic vibration detection results need to be fused with the visual inspection results (e.g., confidence-weighted fusion and D-based fusion). The integration of S-evidence theory aims to fully utilize the spatial positioning advantages and intuitiveness of vision while ensuring independent detection capabilities under extreme conditions.

[0136] In some embodiments of this application, under a confidence-weighted fusion implementation, the excavator's bucket tooth anomaly detection is performed based on a difference index, including: Map the difference index to the confidence level of the acoustic and vibration anomalies of the bucket teeth; Based on images of excavator bucket teeth, determine the confidence level of visual anomalies in the bucket teeth; The confidence scores for acoustic and vibrational anomalies and visual anomalies are fused to obtain the fused anomaly confidence score. Based on the difference indicators and the fused anomaly confidence level, the health status level of the excavator's bucket teeth is determined, and when the bucket tooth health status level indicates that the excavator has a bucket tooth abnormality, the alarm strategy corresponding to the bucket tooth health status level is triggered.

[0137] The aforementioned bucket tooth image can refer to an image that includes the entire leading edge of the bucket and all the bucket tooth areas.

[0138] Let the anomaly confidence level of the acoustic-vibration channel be... The anomaly confidence level of the visual channel is The confidence level for acoustic anomalies can be determined based on a difference index (e.g., Mahalanobis distance). It is obtained through a monotonic function mapping, for example:

[0139] in, The scaling parameter, used to control the curve shape, controls the rate of increase and sensitivity of the difference index (i.e., the degree of acoustic and vibration anomaly) when mapped to anomaly confidence level. The curve shape can refer to the mapping curve used when converting the difference index into acoustic and vibration anomaly confidence level.

[0140] In some embodiments, the confidence level for visual anomalies can be determined jointly by the anomaly probability output by the visual detection model and the image quality assessment result. For example, if the visual detection model determines a high probability of missing teeth, and the image is clear, has normal brightness, and minimal occlusion, then the confidence level for visual anomalies can be set relatively high. Conversely, if the visual detection model also gives a high probability of anomalies, but the image is dark, blurry, mostly obscured by mud, or the lens self-inspection reveals severe contamination, then this high probability itself may not be reliable, and the confidence level for visual anomalies should be appropriately lowered. As another example, if the visual detection model outputs a moderate probability of anomalies, but the image quality is good, the teeth boundaries are clear, and the entire row of teeth is highly visible, then the system can still consider this result to have some reference value and give a moderate level of confidence level for visual anomalies. In actual implementation, the anomaly probability output by the visual detection model is usually used as the base value, and then adjusted upwards or downwards based on image clarity, brightness, occlusion ratio, visible teeth ratio, and camera self-inspection results to finally obtain the confidence level for visual anomalies.

[0141] In some embodiments, visual reliability weights can be defined. Acoustic Reliability Weight The algorithm is dynamically adjusted based on the current image's sharpness, brightness, occlusion status, and self-check results. Based on this, the anomaly confidence level after fusion is... The calculation formula is as follows:

[0142] Under conditions of good lighting and relatively clean bucket teeth surfaces, Close to or slightly higher The system relies more on visual judgment, while using acoustic vibration results for cross-validation; however, in situations where the system is heavily encased in mud, completely dark, or has extremely thick dust, resulting in very low visual quality scores, the system may fail to deliver the desired results. Significantly reduced, When the sound becomes dominant, the system automatically switches to a sound and vibration-dominated decision-making mode.

[0143] In this embodiment, the reliability weights of vision and acoustic vibration are dynamically adjusted by image quality assessment and working condition information, so that the system can make full use of the spatial location information and human readability provided by the image when the vision conditions are good, and automatically switch to acoustic vibration as the decision-making body when the vision conditions are poor or fail, thereby forming a complementary multimodal bucket tooth monitoring architecture.

[0144] In some embodiments, the controller may base its decisions on the fused anomaly confidence level. And difference indicators (e.g., Mahalanobis distance) Define multi-level alarm strategies. For example, when and If the system is deemed to be in serious abnormality, an audible and visual alarm will be triggered immediately, and it is recommended to stop the machine immediately for inspection; when If the event is deemed suspicious and abnormal, a prompt will be displayed on the driver's cab interface, and the event will be recorded in the log; when If the result is deemed normal, it is only used for fingerprint database updates. To reduce the impact of occasional false alarms, the system can employ a cumulative judgment strategy within a sliding time window, for example, triggering a formal alarm only after multiple consecutive detections exceeding a threshold within several minutes.

[0145] In this embodiment, the multi-sensor channel attention fusion and visual vibration multimodal confidence fusion mechanism enable the system to utilize the intuitive spatial positioning and auxiliary verification provided by the image when visual conditions are good, and automatically increase the sound vibration decision weight when visual conditions are poor, avoiding misjudgment and missed judgment caused by relying on a single mode, and the overall false alarm rate can be controlled at a low level.

[0146] In some embodiments of this application, based on D Under the fusion implementation of S-evidence theory, acoustic vibration detection results and visual detection results can be constructed into two independent evidence bodies, using D... The S-evidence theory is integrated. Specifically, based on difference indicators, excavator bucket tooth anomaly detection is performed, including: Based on the difference index, the probability of abnormality of the bucket teeth detected by acoustic vibration is determined; The reliability factor of acoustic vibration detection is determined based on multidimensional feature vectors. Based on the probability of abnormality of bucket teeth and the reliability factor of acoustic vibration detection, the basic probability assignment of acoustic vibration evidence is determined. Based on images of excavator bucket teeth, the probability of bucket tooth anomalies detected by visual inspection and the reliability factor of visual inspection are determined. Based on the probability of abnormal teeth in visual detection and the reliability factor of visual detection, the basic probability assignment of visual evidence is determined. Based on the basic probability assignments of acoustic and vibration evidence and visual evidence, abnormal detection of bucket teeth is performed on excavators.

[0147] Among them, the probability representation of bucket tooth anomalies by acoustic vibration detection is based on difference indicators to determine how likely the excavator is to have bucket tooth anomalies. The probability representation of bucket tooth anomalies by visual detection is based on visual detection units to determine how likely the excavator is to have bucket tooth anomalies.

[0148] The reliability factor characterizes whether the judgment of bucket tooth anomalies is trustworthy. The reliability factor is calculated by comprehensively considering several quality indicators and normalized to a range of 0 to 1. For example, on the acoustic and vibration side, the signal-to-noise ratio, whether the effective operating condition triggering conditions are met, channel consistency, and historical stability can be calculated; on the visual side, brightness, image clarity, occlusion ratio, visible bucket tooth ratio, and lens self-inspection results can be calculated. These indicators are then weighted and combined into a comprehensive quality score. This comprehensive quality score can be used as the reliability factor of the current detection result, or it can be obtained after a simple mapping.

[0149] In this embodiment, with To identify the frame, representing an uncertain state, Indicates a normal state. Indicates an abnormal state. The abnormal probability is displayed in the vibration detection result output. With reliability factor Assigning basic probabilities:

[0150] This represents the basic probability assignment of the acoustic and vibration evidence to the conclusion that "the excavator has abnormal bucket teeth". This represents the basic probability assignment of the acoustic and vibration evidence to the conclusion that "the excavator has abnormal bucket teeth". This represents the basic probability assignment for the acoustic vibration evidence body whose state cannot be clearly determined at this time.

[0151] Anomaly probability output by the visual detection unit With reliability factor constitute:

[0152] This represents the basic probability assignment of the visual evidence to the conclusion that "the excavator has abnormal bucket teeth". This represents the basic probability assignment of the visual evidence to the conclusion that "the excavator has abnormal bucket teeth". This represents the basic probability assignment for visual evidence where the state cannot be clearly determined at this time.

[0153] In this embodiment, the DS evidence theory can not only integrate support levels but also quantify "uncertain" information and conflicting evidence. By constructing basic probability assignments, it can clearly distinguish between three states: "normal," "abnormal," and "uncertain." This is particularly effective in handling complex situations such as fluctuations in sensor reliability and conflicting conclusions from two sources of evidence, making the final decision more robust and credible when faced with incomplete or conflicting information.

[0154] In some embodiments of this application, based on the basic probability assignments of acoustic and vibrational evidence and visual evidence, anomaly detection of bucket teeth in excavators is performed, including: The Dempster combination rule is used to fuse the basic probability assignments of acoustic and vibrational evidence and visual evidence to obtain the fused basic probability assignments. Multiply the fused basic probability assignment of the uncertain state in the fused basic probability assignment by a preset value to obtain the target value; The final abnormal probability is obtained by adding the target value to the fused basic probability assignment of the abnormal state in the fused basic probability assignment. Alternatively, subtract the target value and the fused basic probability assignment of the normal state from 1 to obtain the final abnormal probability. If the final anomaly probability is greater than the anomaly probability threshold, then it is determined that the excavator has an anomaly in the bucket teeth; If the final anomaly probability is less than or equal to the anomaly probability threshold, then it is determined that the excavator does not have any bucket tooth anomalies.

[0155] Optionally, preset values ​​and anomaly probability thresholds can be set according to actual needs or empirical values. For example, the preset value is 0.5, and the anomaly probability threshold is 0.7.

[0156] In this embodiment, the final anomaly probability can be calculated using either of the following two formulas:

[0157]

[0158] in, For the final anomaly probability, This is a preset value. Assign values ​​to the basic probabilities after the fusion of uncertain states; Assign values ​​to the base probabilities after merging abnormal states.

[0159] In this embodiment, by reasonably allocating or transforming the remaining "uncertainty" portion after fusion into support for "abnormal" or "normal" propositions, a scalarized final anomaly probability that can be directly compared with a threshold can be obtained. This final anomaly probability is then compared with the anomaly probability threshold to achieve the detection of bucket tooth anomalies.

[0160] In this embodiment, based on D The integration approach of the S-evidence theory has advantages when more rigorous processing of conflicting information is required.

[0161] In some embodiments, during long-term operation, when multiple anomaly detection results are continuously determined to be normal and there are no manual alarm records, the new feature vector can be fused into the fingerprint database in an exponentially weighted manner to update the database. and This allows the supervised model to adapt to natural wear and tear on equipment and long-term environmental changes without the need for frequent retraining.

[0162] In this embodiment of the application, reliable monitoring of bucket tooth anomalies under extreme working conditions is achieved by using acoustic-vibration fusion frequency domain fingerprint modeling, working condition adaptive triggering, and multimodal confidence fusion.

[0163] In this embodiment, frequency domain feature extraction and lightweight algorithm design make the solution suitable for deployment on in-vehicle edge computing units, enabling real-time inference at the edge. The latency for a single detection can be controlled to the tens to hundreds of milliseconds, meeting the requirements for online monitoring and real-time alarms. The overall architecture of this solution is designed for in-vehicle edge computing environments, employing condition-triggered acquisition, lightweight feature extraction, and efficient judgment algorithms. This reduces data processing and communication overhead while meeting the latency requirements for online real-time monitoring.

[0164] In this embodiment, the changes to the original excavator hydraulic system and electrical control logic are minor. The upgrade can be achieved mainly by adding sound and vibration sensors and embedded processing units. It is suitable for gradual retrofitting of existing vehicles and can be centrally managed and predictively maintained through a remote monitoring platform.

[0165] In this embodiment, a fingerprint database of full-tooth states is constructed according to working condition categories, and statistical methods such as Mahalanobis distance are used for unsupervised anomaly detection. At the same time, an online update mechanism is introduced so that the fingerprint database can be gradually adjusted with the natural wear and tear of the equipment and changes in working conditions, which solves the long-tail distribution problem caused by the scarcity and diverse morphology of bad tooth samples in actual engineering.

[0166] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0167] Corresponding to the bucket tooth anomaly detection method described in the above embodiments, Figure 5 A schematic diagram of the structure of the bucket tooth abnormality detection device provided in the embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown.

[0168] Reference Figure 5 The device includes: The signal acquisition module 501 is used to acquire the acoustic vibration timing signal acquired by the acoustic vibration acquisition module. The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the boom or stick of the excavator. The acoustic vibration timing signal includes the acoustic vibration signal at the current moment and at least one acoustic vibration signal acquired at a time earlier than the current moment. The feature construction module 502 is used to construct a corresponding multidimensional feature vector based on the acoustic vibration time-series signal; The difference calculation module 503 is used to calculate the difference index between the multidimensional feature vector and the fingerprint database. The difference index reflects the difference between the multidimensional feature vector and the fingerprint database, which includes standardized feature vectors of full-tooth state samples. The anomaly detection module 504 is used to perform bucket tooth anomaly detection on the excavator based on the difference index.

[0169] In some embodiments, the feature construction module 502 includes: The transformation unit is used to perform a short-time Fourier transform on the acoustic vibration timing signal to obtain the time-frequency distribution; The decomposition unit is used to perform multi-level wavelet packet decomposition on the acoustic vibration time-series signal and calculate the energy ratio characteristics of adjacent frequency bands. The construction unit is used to construct the multidimensional feature vector based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic-vibration timing signal.

[0170] In some embodiments, the building unit is specifically used for: Based on the aforementioned time-frequency distribution, the average power spectral density at each frequency point is calculated; The average power spectral density of each frequency point is normalized to the energy spectral density to obtain the energy spectral density of each frequency point. A specific frequency band is divided into multiple characteristic frequency bands, and the specific frequency band is a frequency range that differs between the full tooth state and the abnormal tooth state. The multidimensional feature vector is constructed based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators.

[0171] In some embodiments, the building unit is specifically used for: For each of the characteristic frequency bands, the energy spectral densities of the corresponding target frequency points are added together to obtain the energy characteristics of the characteristic frequency band, and each target frequency point is a frequency point located within the characteristic frequency band among the frequency points. The frequency index corresponding to the maximum value of the energy spectral density in the characteristic frequency band is determined as the real-time main peak frequency of the characteristic frequency band; Calculate the spectral centroid based on the average power spectral density and the frequency index of each frequency point; Spectral broadening is calculated based on the spectral centroid, the average power spectral density of each frequency point, and the frequency index of each frequency point. The multidimensional feature vector is constructed based on the energy characteristics of multiple characteristic frequency bands, the spectral centroid, the spectral broadening, the energy ratio characteristics of adjacent frequency bands, the real-time main peak frequencies of multiple characteristic frequency bands, and the time-domain statistical indicators of the acoustic vibration time-series signal.

[0172] In some embodiments, the building unit is further configured to: A key feature frequency band is determined from multiple feature frequency bands, wherein the key feature frequency band is the frequency range in which the difference is greatest between the full tooth state and the abnormal tooth state; Calculate the offset corresponding to the key feature frequency band, where the offset corresponding to the key feature frequency band is the offset between the real-time main peak frequency of the key feature frequency band and the average of the standard main peak frequencies of the key feature frequency band in the fingerprint database. The difference calculation module 503 is specifically used for: If the difference index is greater than the difference threshold, and the offset corresponding to the key feature frequency band is greater than the offset threshold, then it is determined that the excavator has an abnormal bucket tooth.

[0173] In some embodiments, the difference calculation module 503 is specifically used for: The difference index is mapped to the acoustic and vibration anomaly confidence level of the bucket teeth; Based on the images of the bucket teeth of the excavator, the confidence level of visual anomalies of the bucket teeth is determined; The acoustic vibration anomaly confidence score and the visual anomaly confidence score are fused to obtain the fused anomaly confidence score. Based on the difference index and the fused anomaly confidence level, the health status level of the excavator's bucket teeth is determined, and when the bucket tooth health status level indicates that the excavator has a bucket tooth abnormality, an alarm strategy corresponding to the bucket tooth health status level is triggered.

[0174] In some embodiments, the difference calculation module 503 is specifically used for: Based on the aforementioned difference index, the probability of abnormality in the acoustic vibration detection of the bucket teeth is determined; Based on the multidimensional feature vector, the reliability factor of the acoustic vibration detection is determined; Based on the probability of abnormality of the bucket teeth detected by the acoustic vibration detection and the reliability factor of the acoustic vibration detection, the basic probability assignment of the acoustic vibration evidence body is determined. Based on the excavator's bucket tooth image, determine the probability of bucket tooth anomalies detected by visual inspection and the reliability factor of the visual inspection. Based on the probability of abnormal teeth detected by the visual inspection and the reliability factor of the visual inspection, the basic probability assignment of the visual evidence body is determined. Based on the basic probability assignments of the acoustic and vibration evidence and the basic probability assignments of the visual evidence, the excavator is subjected to bucket tooth anomaly detection.

[0175] In some embodiments, the difference calculation module 503 is specifically used for: The Dempster combination rule is used to fuse the basic probability assignments of the acoustic and vibrational evidence and the visual evidence to obtain the fused basic probability assignments. Multiply the fused basic probability assignment of the uncertain state in the fused basic probability assignment by a preset value to obtain the target value; The target value is added to the fused basic probability assignment of the abnormal state in the fused basic probability assignment to obtain the final abnormal probability. Alternatively, subtract the target value and the fused basic probability assignment of the normal state from 1 to obtain the final abnormal probability. If the final anomaly probability is greater than the anomaly probability threshold, then it is determined that the excavator has an abnormal bucket tooth.

[0176] In some embodiments, the difference calculation module 503 is specifically used for: When there are multiple acquisition channels for the acoustic vibration timing signal, the attention weights of the multiple acquisition channels are determined. Based on the attention weights of the multiple acquisition channels, the multidimensional feature vectors corresponding to the acoustic and vibration time-series signals of the multiple acquisition channels are weighted and fused to obtain the fused feature vector; Calculate the difference index between the fused feature vector and the fingerprint database.

[0177] In some embodiments, the signal acquisition module 501 is specifically used for: Under the target working condition, the acoustic vibration timing signal collected by the acoustic vibration acquisition module is acquired. The target working condition is any one of the following: the moment the bucket touches the ground, the stable cutting stage, and the no-load lifting stage. The calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Calculate the difference index between the multidimensional feature vector and the target standardized feature vector in the fingerprint database, wherein the target standardized feature vector includes the standardized feature vector of the full-tooth state sample under the target working condition.

[0178] In some embodiments, the above-described apparatus further includes a working condition identification module; the working condition identification module is used for: Monitor the rate of change of Z-axis acceleration at the boom end and the rate of change of pressure in the bucket cylinder; If the rate of change of the Z-axis acceleration at the end of the boom is greater than the acceleration rate of change threshold, and the rate of change of the pressure in the bucket cylinder is greater than the pressure rate of change threshold, then it is determined that the moment the bucket touches the ground has been detected.

[0179] In some embodiments, the difference calculation module 503 is specifically used for: When there are multiple sets of standardized feature vectors for the full-tooth state samples, calculate the mean vector and covariance matrix of the standardized feature vectors for the multiple sets of full-tooth state samples. Based on the multidimensional feature vector, the mean vector, and the covariance matrix, the distance between the multidimensional feature vector and the fingerprint database is calculated, and the distance is determined as the difference index.

[0180] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0181] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 6 of this embodiment includes: at least one processor 60 ( Figure 6 (Only one is shown in the diagram), memory 61, and computer program 62 stored in said memory 61 and executable on said at least one processor 60, wherein said processor 60 executes said computer program 62 to implement the steps in any of the above method embodiments.

[0182] The electronic device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 6 and does not constitute a limitation on electronic device 6. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0183] The processor 60 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0184] In some embodiments, the memory 61 may be an internal storage unit of the electronic device 6, such as a hard disk or memory of the electronic device 6. In other embodiments, the memory 61 may be an external storage device of the electronic device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 6. Furthermore, the memory 61 may include both internal and external storage units of the electronic device 6. The memory 61 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 61 can also be used to temporarily store data that has been output or will be output.

[0185] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0186] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a device / electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0187] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0188] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0189] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0190] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0191] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method of detecting an abnormality of a bucket tooth, characterized by, include: Acquire acoustic vibration timing signals collected by the acoustic vibration acquisition module. The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the boom or stick of the excavator. The acoustic vibration timing signals include the acoustic vibration signal at the current moment and at least one acoustic vibration signal with an acquisition time earlier than the current moment. Based on the acoustic vibration time-series signal, a corresponding multidimensional feature vector is constructed; Calculate the difference index between the multidimensional feature vector and the fingerprint database. The difference index reflects the difference between the multidimensional feature vector and the fingerprint database, which includes standardized feature vectors of samples with full teeth. Based on the aforementioned difference indicators, the excavator is subjected to bucket tooth anomaly detection.

2. The bucket tooth abnormality detection method according to claim 1, characterized by, The construction of a corresponding multidimensional feature vector based on the acoustic vibration time-series signal includes: Perform a short-time Fourier transform on the acoustic vibration time-series signal to obtain the time-frequency distribution; The acoustic vibration time-series signal is subjected to multi-layer wavelet packet decomposition to calculate the energy ratio characteristics of adjacent frequency bands; Based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic-vibration time-series signal, the multidimensional feature vector is constructed.

3. The bucket tooth abnormality detection method according to claim 2, characterized by, The construction of the multidimensional feature vector based on the time-frequency distribution, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators of the acoustic vibration time-series signal includes: Based on the aforementioned time-frequency distribution, the average power spectral density at each frequency point is calculated; The average power spectral density of each frequency point is normalized to the energy spectral density to obtain the energy spectral density of each frequency point. A specific frequency band is divided into multiple characteristic frequency bands, and the specific frequency band is a frequency range that differs between the full tooth state and the abnormal tooth state. The multidimensional feature vector is constructed based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators.

4. The bucket tooth abnormality detection method according to claim 3, characterized by, The construction of the multidimensional feature vector based on multiple characteristic frequency bands, the average power spectral density of each frequency point, the energy spectral density of each frequency point, the energy ratio characteristics of adjacent frequency bands, and the time-domain statistical indicators includes: For each of the characteristic frequency bands, the energy spectral densities of the corresponding target frequency points are added together to obtain the energy characteristics of the characteristic frequency band, and each target frequency point is a frequency point located within the characteristic frequency band among the frequency points. The frequency index corresponding to the maximum value of the energy spectral density in the characteristic frequency band is determined as the real-time main peak frequency of the characteristic frequency band; Calculate the spectral centroid based on the average power spectral density and the frequency index of each frequency point; Spectral broadening is calculated based on the spectral centroid, the average power spectral density of each frequency point, and the frequency index of each frequency point. The multidimensional feature vector is constructed based on the energy characteristics of multiple characteristic frequency bands, the spectral centroid, the spectral broadening, the energy ratio characteristics of adjacent frequency bands, the real-time main peak frequencies of multiple characteristic frequency bands, and the time-domain statistical indicators of the acoustic vibration time-series signal.

5. The bucket tooth abnormality detection method according to claim 4, characterized by, Before performing bucket tooth anomaly detection on the excavator based on the aforementioned difference index, the method further includes: A key feature frequency band is determined from multiple feature frequency bands, wherein the key feature frequency band is the frequency range in which the difference is greatest between the full tooth state and the abnormal tooth state; Calculate the offset corresponding to the key feature frequency band, where the offset corresponding to the key feature frequency band is the offset between the real-time main peak frequency of the key feature frequency band and the average of the standard main peak frequencies of the key feature frequency band in the fingerprint database. The step of detecting bucket tooth anomalies in the excavator based on the difference index includes: If the difference index is greater than the difference threshold, and the offset corresponding to the key feature frequency band is greater than the offset threshold, then it is determined that the excavator has an abnormal bucket tooth.

6. The bucket tooth abnormality detection method according to any one of claims 1 to 4, characterized by The step of detecting bucket tooth anomalies in the excavator based on the difference index includes: The difference index is mapped to the acoustic and vibration anomaly confidence level of the bucket teeth; Based on the images of the bucket teeth of the excavator, the confidence level of visual anomalies of the bucket teeth is determined; The acoustic vibration anomaly confidence score and the visual anomaly confidence score are fused to obtain the fused anomaly confidence score. Based on the difference index and the fused anomaly confidence level, the health status level of the excavator's bucket teeth is determined, and when the bucket tooth health status level indicates that the excavator has a bucket tooth abnormality, an alarm strategy corresponding to the bucket tooth health status level is triggered.

7. The bucket tooth abnormality detection method according to any one of claims 1 to 4, characterized by, The step of detecting bucket tooth anomalies in the excavator based on the difference index includes: Based on the aforementioned difference index, the probability of abnormality in the acoustic vibration detection of the bucket teeth is determined; Based on the multidimensional feature vector, the reliability factor of the acoustic vibration detection is determined; Based on the probability of abnormality of the bucket teeth detected by the acoustic vibration detection and the reliability factor of the acoustic vibration detection, the basic probability assignment of the acoustic vibration evidence body is determined. Based on the excavator's bucket tooth image, determine the probability of bucket tooth anomalies detected by visual inspection and the reliability factor of the visual inspection. Based on the probability of abnormal teeth detected by the visual inspection and the reliability factor of the visual inspection, the basic probability assignment of the visual evidence body is determined. Based on the basic probability assignments of the acoustic and vibration evidence and the basic probability assignments of the visual evidence, the excavator is subjected to bucket tooth anomaly detection.

8. The bucket tooth anomaly detection method according to claim 7, characterized by, The method of detecting bucket tooth anomalies in the excavator based on the basic probability assignments of the acoustic and vibration evidence and the visual evidence includes: The Dempster combination rule is used to fuse the basic probability assignments of the acoustic and vibrational evidence and the visual evidence to obtain the fused basic probability assignments. Multiply the fused basic probability assignment of the uncertain state in the fused basic probability assignment by a preset value to obtain the target value; The target value is added to the fused basic probability assignment of the abnormal state in the fused basic probability assignment to obtain the final abnormal probability. Alternatively, subtract the target value and the fused basic probability assignment of the normal state from 1 to obtain the final abnormal probability. If the final anomaly probability is greater than the anomaly probability threshold, then it is determined that the excavator has an abnormal bucket tooth.

9. The method for detecting abnormal bucket teeth according to any one of claims 1 to 5, characterized in that, When there are multiple acquisition channels for the acoustic vibration time-series signal, the calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Determine the attention weights for multiple acquisition channels; Based on the attention weights of the multiple acquisition channels, the multidimensional feature vectors corresponding to the acoustic and vibration time-series signals of the multiple acquisition channels are weighted and fused to obtain the fused feature vector; Calculate the difference index between the fused feature vector and the fingerprint database.

10. The method for detecting abnormal bucket teeth according to any one of claims 1 to 5, characterized in that, The acquisition of the acoustic vibration time-series signal acquired by the acoustic vibration acquisition module includes: Under the target working condition, the acoustic vibration timing signal collected by the acoustic vibration acquisition module is acquired. The target working condition is any one of the following: the moment the bucket touches the ground, the stable cutting stage, and the no-load lifting stage. The calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: Calculate the difference index between the multidimensional feature vector and the target standardized feature vector in the fingerprint database, wherein the target standardized feature vector includes the standardized feature vector of the full-tooth state sample under the target working condition.

11. The method for detecting abnormal bucket teeth according to claim 10, characterized in that, The method for identifying the moment the bucket touches the ground includes: Monitor the rate of change of Z-axis acceleration at the boom end and the rate of change of pressure in the bucket cylinder; If the rate of change of the Z-axis acceleration at the end of the boom is greater than the acceleration rate of change threshold, and the rate of change of the pressure in the bucket cylinder is greater than the pressure rate of change threshold, then it is determined that the moment the bucket touches the ground has been detected.

12. The method for detecting abnormal bucket teeth according to any one of claims 1 to 5, characterized in that, The calculation of the difference index between the multidimensional feature vector and the fingerprint database includes: When there are multiple sets of standardized feature vectors for the full-tooth state samples, calculate the mean vector and covariance matrix of the standardized feature vectors for the multiple sets of full-tooth state samples. Based on the multidimensional feature vector, the mean vector, and the covariance matrix, the distance between the multidimensional feature vector and the fingerprint database is calculated, and the distance is determined as the difference index.

13. A device for detecting abnormal bucket teeth, characterized in that, include: The signal acquisition module is used to acquire the acoustic vibration timing signal acquired by the acoustic vibration acquisition module. The acoustic vibration acquisition module is installed on a rigid structure near the bucket side of the boom or stick of the excavator. The acoustic vibration timing signal includes the acoustic vibration signal at the current moment and at least one acoustic vibration signal acquired at a time earlier than the current moment. The feature construction module is used to construct a corresponding multidimensional feature vector based on the acoustic vibration time-series signal; The difference calculation module is used to calculate the difference index between the multidimensional feature vector and the fingerprint database. The difference index reflects the difference between the multidimensional feature vector and the fingerprint database, which includes standardized feature vectors of full-tooth state samples. An anomaly detection module is used to detect bucket tooth anomalies in the excavator based on the difference indicators.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it causes the electronic device to implement the bucket tooth abnormality detection method as described in any one of claims 1 to 12.

15. A computer program product, characterized in that, The method includes a computer program, which, when run, causes the method for detecting abnormal teeth as described in any one of claims 1 to 12 to be performed.

Images