Mobile fault detection method and system for dynamic equipment

Abstract

An embodiment of the present invention provides a mobile fault detection method and system for dynamic equipment, belonging to the technical field of fault detection. The method includes the steps of: patrolling and collecting detection signals within the equipment installation area; classifying the detection signals and selecting a pre-defined fault diagnosis model based on the classification result; extracting features from the detection signals to obtain feature information; training the selected fault diagnosis model based on the feature information; and identifying faulty equipment based on the training result. The solution of the present invention realizes mobile detection of dynamic equipment, is compatible with fault detection methods for various signal types, and fully considers the influence of changes in the spatial location of signal collection on signal features, thereby improving the accuracy of fault detection.

Description

[Technical Field]

[0001] The present invention relates to the technical field of fault detection, and particularly to a mobile fault detection method for dynamic equipment and a mobile fault detection system for dynamic equipment. [Background technology]

[0002] Dynamic equipment is widely used in enterprise production and is the core power equipment of enterprise production, such as motor equipment and gear equipment. Since the stability of these equipment is directly related to the stability of the production process, it is essential to detect the operating status of dynamic equipment during the production process. Traditional dynamic equipment fault detection techniques mainly use manual patrol inspections to collect operating vibration status signals of dynamic equipment, and then perform fault detection of dynamic equipment based on the collected vibration signals. In addition to manual patrol inspections, there is currently a method of deploying online detection systems to detect dynamic equipment faults. However, this method requires the deployment of a large amount of hardware equipment, and in order to achieve wide-area detection, even more hardware equipment must be deployed, resulting in high system construction costs. As can be seen, traditional dynamic equipment fault detection mainly combines two methods: fixed-point vibration signal detection and manual fault detection. However, for factories with very large equipment installation areas, this method significantly increases the labor costs and equipment investment costs for dynamic equipment fault detection, and it is difficult to ensure the accuracy of vibration signal fault detection, so there are also major problems with the fault detection accuracy of traditional solutions. To address the problems of traditional dynamic equipment fault detection methods, such as high implementation costs and low accuracy, a new method for dynamic equipment fault detection is needed. Summary of the Invention [Problem to be solved by the invention]

[0003] The embodiments of the present invention aim to at least solve the problem that conventional solutions cannot realize accurate identification of faults based on audio signals generated in dynamic equipment by providing a method and system for mobile fault detection of dynamic equipment. [Means for solving the problem]

[0004] To achieve the above object, a first aspect of the present invention provides a fault detection method for dynamic equipment, the method including: a step of circulating and collecting detection signals within an equipment installation area; a step of classifying the detection signals and selecting a predetermined fault diagnosis model based on the classification result; a step of extracting features from the detection signals to obtain feature information; a step of training the selected fault diagnosis model based on the feature information; and a step of identifying faulty equipment based on the training result.

[0005] Optionally, said detection signals include acoustic and / or vibration signals generated in the target equipment, location information at the current collection time and operating parameters of the target equipment.

[0006] Optionally, the step of cyclically collecting detection signals within the equipment installation area includes a step of determining a filtering limit margin of the target equipment based on the location information at the current collection time and the operating parameters of the target equipment, a step of determining an actual filtering limit of the target equipment based on the filtering limit margin and a predetermined basic filtering limit, and a step of collecting detection signals of the target equipment based on the actual filtering limit.

[0007] Optionally, the step of determining a filtering limit margin of the target equipment based on the location information at the current collection time and the operating parameters of the target equipment includes the steps of determining distance information between the location at the current collection time and the target equipment based on the location information at the current collection time, determining detection signal prediction information of the target equipment based on the operating parameters of the target equipment, and determining a filtering limit margin of the target equipment based on the distance information and the detection signal prediction information, wherein the detection signal prediction information includes predicted values ​​of one or more of the amplitude, frequency, and wavelength of the detection signal.

[0008] Optionally, the step of determining a filtering limit margin of the target equipment based on the distance information and the detection signal prediction information includes the steps of determining an attenuation rule for the detection signal prediction information based on the distance information, determining detection signal prediction information for the position of the current collection time based on the attenuation rule, and determining a filtering limit margin of the target equipment based on the detection signal prediction information for the position of the current collection time.

[0009] Optionally, the step of classifying the detection signals includes the steps of respectively determining the relationship between the currently collected acoustic signal and vibration signal and a predetermined minimum acoustic signal intensity and a predetermined vibration signal intensity, and classifying the detection signals based on the determination result; if the currently collected acoustic signal is smaller than the predetermined minimum acoustic signal intensity but the vibration signal is equal to or greater than the predetermined vibration signal intensity, the classification result of the currently collected detection signal is a vibration signal; if the currently collected acoustic signal is equal to or greater than the predetermined minimum acoustic signal intensity but the vibration signal is smaller than the predetermined vibration signal intensity, the classification result of the currently collected detection signal is an acoustic signal; if the currently collected acoustic signal is equal to or greater than the predetermined minimum acoustic signal intensity and the vibration signal is equal to or greater than the predetermined vibration signal intensity, the classification result of the currently collected detection signal is a fusion signal of the vibration signal and the acoustic signal; if the currently collected acoustic signal is smaller than the predetermined minimum acoustic signal intensity and the vibration signal is smaller than the predetermined vibration signal intensity, the currently collected signal does not meet requirements, so adjusting the signal collection position and re-collecting the detection signal.

[0010] Optionally, the preset fault diagnosis models include a fault diagnosis model based on acoustic signals, a fault diagnosis model based on vibration signals, and a fusion fault diagnosis model based on acoustic signals and vibration signals.

[0011] Optionally, the step of selecting a preset fault diagnosis model based on the classification result includes the steps of: if the classification result is a vibration signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on the vibration signal; if the classification result is an acoustic signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on the acoustic signal; and if the classification result is a fusion signal, the correspondingly selected preset fault diagnosis model is a fusion fault diagnosis model based on the acoustic signal and the vibration signal.

[0012] Optionally, when the current model is a fusion fault diagnosis model based on acoustic signals and vibration signals, the step of performing feature extraction on the detection signals and obtaining feature information includes the steps of performing feature extraction on the detection signals based on a preset EMD decomposition algorithm to perform feature fusion and obtain a fusion feature vector set; and performing fault word replacement on each fusion feature vector in the fusion feature vector set based on a preset fusion bag-of-words model to obtain a word evaluation set as feature information.

[0013] Optionally, the step of performing feature extraction on the detection signals based on a preset EMD decomposition algorithm to perform feature fusion and obtain a fused feature vector set includes the steps of: performing EMD decomposition on the synchronously collected acoustic signals and vibration signals to obtain multiple signal components for each signal; performing HHT transformation on each signal component to obtain the instantaneous frequency and instantaneous amplitude of each signal component; comparing the instantaneous frequency of the signal component of the acoustic signal with the instantaneous frequency of the signal component of the vibration signal, adding up the instantaneous amplitudes corresponding to the same points of the instantaneous frequencies of both signals to obtain multiple fused signals, and constructing and obtaining corresponding time-frequency matrices as a fused feature vector set.

[0014] Optionally, the preset fusion bag-of-words model includes: a fault word substitution model used to perform fault word substitution based on the fusion feature vector; and a preset codebook used to record the correspondence between the word frequency vector and the fault cause.

[0015] Optionally, the method further includes a step of constructing a fault word substitution model, which includes the steps of collecting detection signals in a fault state in historical data of dynamic equipment, or collecting detection signals in an operating state of dynamic equipment labeled as being in a fault state, as basic data; performing feature extraction on the basic data to perform feature fusion, and obtaining a fused feature vector training set; and clustering the fused feature vector training set according to a preset clustering algorithm, obtaining a plurality of clustering centers, and performing fault word labeling for each clustering center, and obtaining a fault word substitution model.

[0016] Optionally, the step of training a selected fault diagnosis model based on the feature information and identifying faulty equipment based on the training result includes the steps of: introducing the word evaluation set and the operating process parameters of the target equipment as input parameters into the fusion fault diagnosis model based on the acoustic signal and the vibration signal; performing statistics on the occurrence frequency of fault words based on a preset clustering algorithm; obtaining a corresponding word frequency vector based on the statistical result of the occurrence frequency of fault words; and identifying the cause of the fault against a preset codebook based on the word frequency vector.

[0017] Optionally, the fault diagnosis model further includes a faulty equipment location identification model, and the step of identifying faulty equipment based on the training result further includes a step of identifying the location of faulty equipment, which includes the steps of cyclically collecting acoustic signals to be identified within a predetermined collection area, recording collection location information for each collection time of the acoustic signals to be identified, and obtaining a correspondence between multiple groups of acoustic signals to be identified and the collection location information; and training the faulty equipment location identification model using all the acoustic signals to be identified and the corresponding collection location information as input parameters, and obtaining location information of the faulty equipment.

[0018] Optionally, the step of circularly collecting the acoustic signals to be identified within the predetermined collection area includes a step of collecting a flow of the acoustic signals to be identified by circularly traveling in a linear manner along any two opposite directions starting from the dynamic equipment until the step reaches a position having a predetermined maximum distance from the dynamic equipment, wherein the collection position of the acoustic signals to be identified is different each time, and the distance between the collection positions of any two adjacent acoustic signals to be identified is the same, and at each collection position of the acoustic signals to be identified, multiple acoustic signals to be identified are collected synchronously in a multi-channel manner.

[0019] Optionally, the step of recording collection position information for each collection time of the acoustic signal to be identified includes the steps of: reading laser navigation data circling the current collection position in response to a collection trigger signal of the acoustic signal; reading first collection position candidate information for the current collection position based on the laser navigation data; reading Beidou positioning information for the current collection position in response to a collection trigger signal of the acoustic signal; reading second collection position candidate information for the current collection position based on the Beidou positioning information; and performing collection position correction based on the first collection position information and the second collection position information to obtain the collection position.

[0020] Optionally, the method further includes the steps of: within each audio frequency band, obtaining sound field distribution map information for each frequency band, comparing the sound field distribution map information for each frequency band with preset standard sound field distribution map information to obtain a sound field distribution map deviation degree for each frequency band; obtaining a total deviation degree matrix based on the sound field distribution map deviation degree for each frequency band; training a preset sound field fault identification model using the total deviation degree matrix as an input parameter to obtain an equipment abnormality result in a current operating state; and comparing the equipment abnormality result obtained based on the sound field fault identification model with the equipment abnormality result obtained based on a fault diagnosis model, and verifying the equipment abnormality result obtained by the current fault diagnosis model.

[0021] A second aspect of the present invention provides a mobile fault detection equipment for dynamic equipment, the equipment including: a signal collection robot, a processing unit, and a training unit; the signal collection robot is used to patrol and collect detection signals within the equipment installation area; the processing unit is used to classify and process the detection signals, select a pre-defined fault diagnosis model based on the classification result, and perform feature extraction on the detection signals to obtain feature information; the training unit is used to train the selected fault diagnosis model based on the feature information, and identify faulty equipment based on the training result.

[0022] Optionally, said detection signals include acoustic and / or vibration signals generated in the target equipment, location information at the current collection time and operating process parameters of the target equipment.

[0023] Optionally, the signal collection robot includes a collection module, a traveling module, and a navigation module, wherein the collection module includes a plurality of acoustic signal collectors arranged in a circular array and used for synchronously collecting acoustic signals generated from the target dynamic equipment, and a laser vibrometer used for emitting a vibration measurement laser to the target dynamic equipment and collecting corresponding vibration signals, wherein the traveling module is used to drive the patrol inspection and movement of the patrol inspection robot, and the navigation module is used for obstacle avoidance during the movement of the patrol inspection robot.

[0024] A third aspect of the present invention provides a mobile fault detection system, said system including a mobile fault detection facility for the dynamic equipment described above.

[0025] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, the instructions, when executed by a computer, causing the computer to perform the mobile fault detection method for dynamic equipment described above. [Effects of the Invention]

[0026] Based on the above technical solutions, the solution of the present invention performs circular signal collection, fully considers the influence of spatial signal collection location on signal characteristics, and synchronizes the recording of signal collection location and the collection of equipment process operating parameters. A corresponding model is selected according to the signal type, thereby improving the accuracy of fault detection. A spatial faulty equipment location solution is proposed to ensure accurate identification of faulty equipment.

[0027] Other features and advantages of the embodiments of the present invention are described in detail in the specific embodiment section below. [Brief explanation of the drawings]

[0028] The drawings are included to provide a further understanding of embodiments of the invention, constitute a part of the specification, and, together with the following specific embodiments, are intended to explain embodiments of the invention but are not intended to be limitations thereon.

[0029] [Figure 1] 2 is a flowchart of a mobile fault detection method for dynamic equipment according to an embodiment of the present invention. [Figure 2] 10 is a flowchart of Example 2 according to one embodiment of the present invention. [Figure 3] 10 is a flowchart of Example 3 according to one embodiment of the present invention. [Figure 4] 1 is a diagram illustrating the configuration of a mobile fault detection system for dynamic equipment according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0030]

[0023] Specific embodiments of the present invention will be described in detail below with reference to the drawings. It should be understood that the specific embodiments described herein are for the purpose of explaining and interpreting the present invention, and are not intended to limit the present invention.

[0031] Smart factories represent a new stage in the informationization of modern factories. Based on a digital factory, they utilize Internet of Things (IoT) and equipment monitoring technologies to enhance information management and services, clearly understand production and sales processes, improve production process control, reduce manual intervention on the production line, collect production line data in real time and accurately, and achieve rational production planning and progress. They integrate environmentally friendly smart measures and emerging technologies such as smart systems to create efficient, energy-saving, environmentally friendly, and user-friendly factories. Smart factories represent the future direction of factory development, and one research area is smart equipment failure monitoring. Only through smart equipment failure monitoring can smart equipment operational stability be monitored. Ensuring the stability of the production process is an important aspect of production control.

[0032] In conventional equipment fault detection technologies, vibration state signals are mainly collected during equipment operation through manual patrol inspections, and then equipment faults are detected based on the collected vibration signals. This method requires high labor costs and is not suited to the current development of smart factories. In addition, in many cases, abnormal vibration states only become apparent when the equipment experiences major failures, and the accuracy of fault identification obtained through this method still has certain problems. In addition to the manual patrol inspection method, there is also a method of deploying online monitoring systems to monitor equipment faults. However, this method requires the deployment of a large amount of hardware equipment, and in order to achieve wide-area monitoring, even more hardware equipment must be deployed, resulting in large system construction costs.

[0033] Dynamic equipment is widely used in production sites and is the driving force behind production. During the operation of dynamic equipment, vibrations generate sound. If the equipment has a fault, the generated sound signal also changes, and different types of faults generate different sound signal types. Therefore, estimating the corresponding fault type based on sound signals is an excellent research direction. By collecting sound signals mechanically and analyzing characteristic signals such as the frequency and intensity of the sound signals, the type of fault currently present in the equipment can be inferred. However, dynamic equipment is built in a variety of ways, with different structures and power output methods, which leads to significant differences in the generated sound signals. Currently, there is no solution that can detect faults in all dynamic equipment using a single sound recognition method. To address the problems of conventional methods, the solution of the present invention proposes a new equipment anomaly identification method based on sound signal monitoring. It develops different sound signal analysis methods for different structural characteristics and power output methods of dynamic equipment, and the system adaptively selects the analysis method based on the collected sound signals. This improves the system's ability to analyze sound signals for different scenarios, improves the accuracy of fault identification, and also improves the intelligence of fault identification methods based on sound signals.

[0034] 1 is a flowchart of a mobile fault detection method for dynamic equipment according to an embodiment of the present invention. As shown in FIG. 1, an embodiment of the present invention provides a mobile fault detection method for dynamic equipment, which includes the following steps:

[0035] Step S10: The detection signals within the equipment installation area are collected cyclically.

[0036] The solution of the present invention detects faults based on acoustic and / or vibration signals. Generally speaking, the solution of the present invention does not require disassembling the dynamic equipment and investigating faults for individual components, but instead performs fault estimation based on characteristic information generated by the dynamic equipment. The advantage of this method is that it does not require intervention in the dynamic equipment's operation, makes information collection easy, and does not affect the normal operation of the equipment. When dynamic equipment operates normally, its operating vibration and generated noise are both small. However, when a fault occurs, such as bearing wear, misalignment, imbalance, mechanical looseness, shaft misalignment, or loose foundation, the designed balance state of the dynamic equipment is disrupted, resulting in unwanted noise such as collision and friction, and corresponding vibration signals. Different fault types generate different acoustic and vibration signals; for example, the frequency of asymmetric vibration of a shaft is lower than the frequency of bearing wear. Based on the mapping relationship between the generated information and the corresponding fault conditions, fault identification of the corresponding dynamic equipment can be performed based on the collected information, i.e., acoustic and vibration signals.

[0037] Based on this, it is necessary to collect acoustic signals and vibration signals when performing patrol signal collection. When collecting acoustic signals, there are many moving equipment in the production site, and these equipments all emit sounds during operation, and there are also other background sounds, such as ambient noise and artificial operation sounds, and these sounds are mixed together. Therefore, the difficulty in realizing the solution of the present invention is how to extract the operation sound of the target moving equipment from the many sounds. Preferably, to solve the problem of extracting the operation sound of the target moving equipment, the solution of the present invention proposes a patrol sound collection method. That is, an audio collection robot is installed in advance, and the robot can move and patrol the production site, and audio is collected during the robot's patrol process.

[0038] In an embodiment of the present invention, the design of a collection robot reduces the number of collection devices required, and by pre-setting the robot's travel route, it can independently complete wide-area sound detection tasks, avoiding the problem of traditional equipment requiring extensive on-site hardware installation and facilitating wide-area sound detection. On the other hand, since the location of the moving equipment is fixed and the sound attenuation along the propagation path is constant, by converting the relative distance between the collection robot and the target moving equipment and combining the corresponding sound propagation law, it is possible to effectively identify the sound emitted by the target moving equipment that conforms to the movement law from among multiple sounds. For example, starting from the target moving equipment, gradually moving away from the moving equipment, the distance value between the target moving equipment and the collection robot is the same as the movement value of the collection robot, and the distance between other diagonal moving equipment and the collection robot needs to be converted diagonally. By estimating the sound attenuation law according to the movement distance of the collection robot, it is possible to easily identify the corresponding target moving equipment from among multiple sounds.

[0039] In another possible embodiment, the collection robot includes multiple audio collection channels, which are distributed in different positions and directions. Assuming that the relative positions of the moving equipment and the collection robot are fixed, it is expected that the audio emitted from only one piece of moving equipment will satisfy the rules for multi-channel audio collection. For example, the position between point A and point B is fixed, the collection robot is located at point A and heading toward point B, and the target moving equipment is located at point B. Assuming that the position and orientation of the multi-channel audio collection module of the collection robot are also fixed and satisfy the rules for audio propagation, the audio from each direction from point B to point A will be unique. Based on this uniqueness, the operating sound of the target moving equipment can be identified.

[0040] According to the above rules, the collection robot first determines the corresponding target dynamic equipment, then determines the specific location of the equipment, and drives the collection robot to patrol the specific location as a target position to the location of the target dynamic equipment. Starting from the dynamic equipment to be identified, the collection robot patrols linearly along any two opposite directions until it reaches a position with a predetermined maximum distance from the dynamic equipment to be identified, collecting a stream of audio information to be identified, where multiple pieces of audio information are synchronously collected in a multi-channel manner at each audio collection position. The collection position of the audio information to be identified is different each time, and the distance between any two adjacent collection positions of the audio information to be identified is the same. At each audio collection position, multiple pieces of audio information are synchronously collected in a multi-channel manner.

[0041] Specifically, for collecting vibration signals, conventional vibration signal collection methods mainly use contact-type vibration signal collection, which requires a contact connection between a sensor that collects vibration signals and the dynamic equipment that generates vibrations, and then the sensor and the dynamic equipment vibrate with the same characteristics, thereby collecting corresponding vibration signals. The technical effect that the solution of the present invention aims to achieve is to collect vibration signals during a patrol process, that is, the signal collection robot is mobile, and if vibration signals were collected based on the conventional method, the position of the collection robot would need to be fixed and it would need to be in contact with the dynamic equipment, which is contrary to the original purpose of the present application.

[0042] Example 1

[0043] To handle non-contact vibration signal collection, the solution of the present invention further proposes another vibration signal collection method, namely, directly collecting the vibration signals of the dynamic equipment by contacting the dynamic equipment. In this embodiment, the collection robot is equipped with a corresponding vibration signal collection manipulator, which can make lap contact with the dynamic equipment and then directly collect the corresponding vibration signals along with the vibration process of the dynamic equipment. Because the collection manipulator directly collects the vibration signals of the dynamic equipment, there is no need for corresponding signal conversion, so the collected vibration signals are more accurate. In the patrol process, the collection robot can temporarily stop when it reaches each dynamic equipment to collect the vibration signals, so there is no significant impact on the essential patrol monitoring of the dynamic equipment.

[0044] To avoid this situation, the solution of the present invention preferably uses a laser vibrometer to collect vibration signals, achieving contactless vibration signal collection. A laser vibrometer is a measuring device that measures the vibration of an object using principles such as the laser Doppler effect and optical heterodyne interference. Compared to conventional sensors such as accelerometers, it has advantages such as long-distance measurement, non-contact, high spatial resolution, short measurement time, wide frequency response, and high velocity resolution. It is widely used in fields such as model modal characteristic analysis, quality detection, online control, structural flaw detection, and health care.

[0045] In one possible embodiment, the collection robot is equipped with an OptoMET digital laser Doppler vibrometer, a high-precision vibration measurement device. This device can accurately measure vibration and acoustic signals, including vibration displacement, velocity, and acceleration, without contact. It boasts ultra-high optical sensitivity and utilizes the company's proprietary UltraDSP digital signal processing technology, enabling it to not only quickly measure the vibrations of simple systems, but also measure extremely challenging systems, including those with high-frequency vibrations, long-distance testing, small amplitudes, high linearity, and high vibration acceleration or velocity. UltraDSP ensures high measurement resolution and accuracy. The OptoMET laser vibrometer has excellent linearity and a wide test frequency band, reaching up to 10 MHz.

[0046] Step S20: The detection signal is classified, and a preset fault diagnosis model is selected based on the classification result.

[0047] Specifically, to train different fault identification models for different operating states and obtain the most accurate fault diagnosis results for the corresponding scenes, the solution of the present invention provides three fault diagnosis models: a fault diagnosis model based on acoustic signals, a fault diagnosis model based on vibration signals, and a fusion fault diagnosis model based on acoustic and vibration signals. The three models correspond to three different operating states, and the operating states are distinguished based on the characteristics of the detected signals. To ensure that the collected signals can be used for processing, the signal strength must meet the requirements, and the solution of the present invention performs corresponding classification processing based on the strength of the detected signals.

[0048] Specifically, the relationship between the currently collected acoustic signal and vibration signal and the preset minimum acoustic signal intensity and the preset vibration signal intensity is respectively determined, and the detection signal is classified based on the determination result. If the currently collected acoustic signal is smaller than the preset minimum acoustic signal intensity but the vibration signal is equal to or greater than the preset vibration signal intensity, the classification result of the currently collected detection signal is a vibration signal. If the currently collected acoustic signal is equal to or greater than the preset minimum acoustic signal intensity but the vibration signal is smaller than the preset vibration signal intensity, the classification result of the currently collected detection signal is a vibration signal and an acoustic signal. If the currently collected acoustic signal is equal to or greater than the preset minimum acoustic signal intensity and the vibration signal is equal to or greater than the preset vibration signal intensity, the classification result of the currently collected detection signal is a fusion signal of a vibration signal and an acoustic signal. If the currently collected acoustic signal is smaller than the preset minimum acoustic signal intensity and the vibration signal is smaller than the preset vibration signal intensity, the currently collected signal does not meet the requirement, so the signal collection position needs to be adjusted and the detection signal needs to be collected again.

[0049] If the classification result is a vibration signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on the vibration signal; if the classification result is an acoustic signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on the acoustic signal; if the classification result is a fusion signal, the correspondingly selected preset fault diagnosis model is a fusion fault diagnosis model based on the acoustic signal and the vibration signal.

[0050] Step S30: Select a corresponding fault diagnosis model based on the classification result.

[0051] Specifically, different fault diagnosis models require different input parameters, and correspondingly, different feature information is required when performing signal processing. Based on this, there are multiple situations when performing fault diagnosis.

[0052] Example 2

[0053] When the current model is a fusion fault diagnosis model based on an acoustic signal and a vibration signal, the step of performing feature extraction on the detection signal and obtaining feature information includes the steps of performing feature extraction on the detection signal based on a preset EMD decomposition algorithm to perform feature fusion and obtain a fusion feature vector set, and performing fault word replacement on each fusion feature vector in the fusion feature vector set based on a preset fusion bag-of-words model to obtain a word evaluation set and use it as feature information.

[0054] Specifically, as shown in FIG. 2, it includes the following steps:

[0055] Step S311: Perform feature extraction and feature fusion on the evaluation information based on a preset EMD decomposition algorithm to obtain a fusion feature vector set.

[0056] Specifically, the solution of the present invention does not simply perform fault identification based on acoustic signals and vibration signals, respectively, and then determine the final fault identification method based on a comparison of the identification results. Although this method can solve the problem of a certain level of identification accuracy, the comparison of the two data sets does not accurately determine whether an error has occurred in either data, and multiple calculations must be performed to eliminate accidental errors. As a result, the system must perform the same judgment steps multiple times, increasing the time required for fault identification. This has a significant impact on the efficiency of the inspection patrol as a solution to fault inspections on production sites.

[0057] On the other hand, due to the background noise in a complex background environment and the interrelated influence of each structural component, the signal-to-noise ratio of the noise signal or vibration signal is low, the feature orientation is unclear, and the accuracy of diagnosis and judgment is low. As a result, fault identification relying solely on acoustic signals or solely on vibration signals has low corresponding accuracy. To avoid this situation, the solution of the present invention establishes a coherence relationship for multi-feature diagnosis through a correlation comparison between noise diagnostic feature information and vibration diagnostic feature information, and establishes a vibration-noise related fusion judgment of faults and diagnostic features.

[0058] Specifically, acoustic signals and vibration signals are essentially the same, and theoretical analysis shows that the fault diagnosis results based on acoustic signals should be consistent with the diagnosis results based on vibration signals. By analyzing the correlation between the two and then performing fusion processing on the collected signals, the accuracy of the identification results can be greatly improved and the robustness of the system can be improved. The solution of the present invention uses the coherence function to discuss the correlation between the two. The coherence function is an actual function that describes the correlation between the input signal and the output signal in the frequency domain of the system, and can measure the causal relationship between the two signals, corresponding to the correlation analysis in the time domain of the system. Specifically, it is as follows:

[0059]

number

[0060] where f is the frequency,

number

number

number

number

number

number

[0061] On the other hand, to further improve the fault identification accuracy of dynamic equipment, a related model of both signals is established, and then corresponding signal fusion is performed. Fault identification is performed based on the fused signal, with identification accuracy much higher than that of fault identification based on either signal alone. To achieve feature fusion, feature extraction is performed based on the acoustic signal and vibration signal, respectively, and then fusion is performed based on the extracted features.

[0062] Preferably, the solution of the present invention performs feature extraction on the evaluation information based on a preset EMD decomposition algorithm to perform feature fusion. Specifically, EMD decomposition is performed on the synchronously collected acoustic signals and vibration signals, each of which obtains multiple signal components, and HHT transformation is performed on each signal component to obtain the instantaneous frequency and instantaneous amplitude of each signal component. The instantaneous frequency of the signal component of the acoustic signal is compared with the instantaneous frequency of the signal component of the vibration signal, and the instantaneous amplitudes corresponding to the same points in the instantaneous frequencies of both signals are added to obtain multiple fused signals, and corresponding time-frequency matrices are constructed and obtained as a fused feature vector set.

[0063] The most notable feature of empirical mode decomposition (EMD) as a time-frequency domain processing method is that it overcomes the problem of non-adaptive basis functions. For unknown signals, decomposition can be initiated directly without prior analysis or research. This method automatically divides signals hierarchically according to several fixed modes, eliminating the need for manual configuration or intervention. EMD decomposition can obtain multiple signal components, i.e., intrinsic mode functions (IMFs). The intrinsic mode components are subject to the following two constraints: first, the number of extreme points and the number of zero-crossing points within the entire data segment must be equal, or the difference must not exceed one. second, at any time, the average value of the upper envelope consisting of local maximum points and the lower envelope consisting of local minimum points must be zero, i.e., the upper and lower envelopes are locally symmetrical with respect to the time axis. This decomposition method can be used to extract features from chaotic acoustic and vibration signals, followed by corresponding feature fusion.

[0064] Specifically, each signal component is first subjected to HHT transformation, which is also a common signal stabilization processing method after using an empirical mode transformation method. It is known from theory that sound is generated by vibration, so there is a certain correlation between the acoustic signal frequency and the vibration frequency. Based on this correlation, the same point of peak frequency of the acoustic signal component and the vibration signal component is found, and then signal amplitudes are superimposed based on this point. This ensures that the two fused signals are generated simultaneously by the same equipment and that the fused signal matches the actual situation of the equipment. Because there are multiple signal components, there are also multiple corresponding fused signals, and a fused feature vector set is constructed based on these fused signals.

[0065] Step S312: According to a preset fusion bag-of-words model, perform fault word replacement on each fusion feature vector in the fusion feature vector set to obtain a word evaluation set.

[0066] Specifically, the constructed fusion feature vector is a representation format of the collected signals, and the representation format needs to be mapped to the corresponding fault state, that is, the fusion feature vector in the fusion feature vector set needs to be mapped to the fault based on the mapping relationship between the fault state and the description information feature, where a pre-constructed fusion bag-of-words model needs to be applied, which can complete the conversion from the feature vector to the fault cause.

[0067] Preferably, the solution of the present invention uses a bag-of-words library model. The bag-of-words library is used to calculate the frequency of each fault word and construct a vector indicating the occurrence of a fault from the word frequency, thereby significantly reducing the amount of computation required for the overall information. The bag-of-words model places all words in a document library into a single bag, without considering grammar and word order issues; that is, each word is independent. Words appearing in each document are mapped to a dictionary. If a word appears, its corresponding position is incremented by 1; otherwise, it is decremented by 0. Finally, the number at a position indicates the number of times the word appears in the document. Because different fusion feature vectors may indicate different fault causes, it is necessary to find the word vector with the highest frequency of occurrence. Generating a corresponding frequency vector based on the frequency relationship can simplify large amounts of data, reduce the amount of data computation, and ensure the accuracy of fault cause identification. The bag-of-words model meets the needs of the solution of the present invention. Collect evaluation information in a fault state from the historical data of dynamic equipment, or collect evaluation information in an operating state of dynamic equipment labeled as being in a fault state, use this as basic data, perform feature extraction on the basic data to perform feature fusion, obtain a fused feature vector training set, cluster the fused feature vector training set according to a preset clustering algorithm to obtain a plurality of clustering centers, perform fault word labeling for each clustering center, and obtain a fault word substitution model, obtain a corresponding word frequency vector training set according to the fault word substitution model and the fused feature vector training set, and build a corresponding codebook for the word frequency vector training set according to the fault causes recorded in the historical data or the labeled equipment fault causes, which is used as a preset codebook.

[0068] In one possible embodiment, k clustering centers are obtained from all feature vectors using the k-means algorithm, and the clustering centers are set as fault words. Then, using the extracted fault words, the frequency of occurrence of all fault words in noise and vibration signals is calculated based on the clustering principle, and fault features are reconstructed based on this.

[0069] Step S313: Perform statistics on the occurrence frequency of fault words based on a preset clustering algorithm, and perform fault diagnosis of the target dynamic equipment based on the statistical results.

[0070] Specifically, after obtaining the word evaluation set, the corresponding word frequency vector can be obtained, and the word frequency vector can be used as a search criterion to search for the corresponding fault cause in the pre-defined codebook, and find the fault state that best matches the current evaluation information, thereby completing the identification of the fault cause of the dynamic equipment.

[0071] Example 3

[0072] Current models are fault diagnosis models based on acoustic or vibration signals.

[0073] Specifically, to facilitate the collection robot's patrol, a corresponding laser navigation module is installed. After determining the target location, an optimal path is planned based on pre-defined map information. Then, during the patrol, forward guidance and obstacle avoidance operations are performed based on the laser navigation data. Attenuation occurs as the sound propagation distance changes, resulting in a decrease in sound intensity and a constant frequency. While certain compensation is performed on the collected sound, to filter out the sounds of other moving equipment, the collected sound features must be combined with spatial location features, thereby extracting and enhancing the target sound in combination with sound propagation rules. In response to a sound information collection trigger signal, the robot reads laser navigation data patrolling the current collection location, reads first collection location candidate information for the current collection location based on the laser navigation data, reads Beidou positioning information for the current collection location based on the Beidou positioning information, and then performs information correction based on the first and second collection location information to obtain collection location information.

[0074] To avoid noise interference, eliminate sudden noise, and improve accuracy, the solution of the present invention collects multiple sets of audio information to be identified and corresponding location information. Enlarging the amount of data reduces errors, and the multiple sets of data are collected from data collected at different patrol positions of the collection robot, while also being collected from audio data synchronously using a multi-array method. That is, one set of location information corresponds to multiple sets of audio information to be identified, and sudden noise interference is avoided.

[0075] Preferably, the audio information to be identified includes operating sound information of the dynamic equipment to be identified and background sound information, and the audio information includes audio intensity information, audio frequency domain feature information, and audio waveform information.

[0076] In one possible embodiment, after identifying the target dynamic facility, the collection robot patrols the location of the target dynamic facility, selects a direction from the dynamic facility that is easy to move in, and performs patrol collection of audio. Each time it moves a certain distance away, it collects audio information once. After reaching a preset distance threshold, it collects one final audio information in its current direction. It then returns to the location of the dynamic facility and patrols along another direction from the originally selected direction, repeating the audio collection operation. For example, it samples audio within a range of 50 cm (i.e., 1 m) in front and behind the facility to be detected. To further improve accuracy, the sampling rate is preferably set to be greater than twice the signal frequency, i.e., fs > 2 × fn. Each second of collected audio signal is recorded as one waveform data, i.e., one second of collected audio is saved as one waveform, and the next second of collected audio is saved as one waveform. The timestamp of the start of each waveform collection and the location stamp of the robot's location at that time are input to the algorithm model, along with the waveform data. Specifically, as shown in Figure 3.

[0077] Step S321: Using all the voice information to be identified and the corresponding collection location information as input parameters, a preset fault identification model is trained, and the fault identification result of the corresponding relationship of each group is obtained.

[0078] Specifically, as described above, the operating state of the dynamic equipment is indicated by voice changes, and after voice information collection and corresponding location information collection are completed, the operating state of the dynamic equipment can be inversely estimated based on the collected information. Before that, a corresponding fault identification model needs to be trained, and the fault identification model can derive whether the dynamic equipment has an operating fault using the voice information and the corresponding location information as input parameters.

[0079] Specifically, the method collects operating sound information from multiple pieces of dynamic equipment labeled as being in a fault state, and also collects background sound information from the non-operating state of the dynamic equipment at the installation site of the multiple pieces of dynamic equipment. The collected operating sound information and background sound information are used as training samples to train a fault identification model. The operating sound information from the dynamic equipment in a fault state can be audio information collected directly from the dynamic equipment in a fault state, or historical information recorded by the dynamic equipment in a historical fault state. The larger the amount of data, the more the fault identification model finally obtained through training will meet the requirements. The collection locations are then correspondingly recorded, and the collection locations are bound to the corresponding fault audio information. The bound information is used as training sample data. When training the model, part of the training sample data can be reserved as test data. That is, in the case of a subsequent initial model, the trained model is tested based on the test data to determine whether its identification results meet the requirements.

[0080] In another possible embodiment, if a fault occurs in the dynamic equipment, manual maintenance is required, thereby preventing the fault from developing into a larger one. Therefore, identifying the corresponding dynamic equipment as being in a fault state can meet the user's needs. However, to further improve the intelligence of the method and reduce the time required for the user to investigate the fault, identifying an operating fault in the dynamic equipment and simultaneously providing the user with a specific fault investigation direction would be of significant significance in improving the efficiency of subsequent fault maintenance. For dynamic equipment labeled as being in a fault state or from historical data, the fault results, i.e., the specific faults that cause changes in voice characteristics, are known. If the relationship between specific fault causes and unique voice characteristics can be found, a certain degree of analysis of the fault cause can be performed based on the collected voice, providing auxiliary information for the user when investigating the fault and improving the efficiency of fault detection. Operation sound information from multiple dynamic equipment labeled as being in a fault state is collected, and each fault type corresponding to the voice information is bound to the sound information. Background sound information from the non-operating state of the dynamic equipment at the installation site of the multiple dynamic equipment is collected. The multiple voice information and multiple background sound information bound to the corresponding fault types are used as training samples to train a fault identification model.

[0081] Preferably, when training the model, it is also necessary to train the model with background sound, thereby ensuring that the model has the function of removing background sound and ensuring that the final extracted sound is from the target dynamic equipment only.

[0082] Step S322: The result of fault identification is calculated as the proportion of the correspondence relationships with faults to the total number of correspondence relationships, and if the proportion is greater than a preset threshold, it is determined that the corresponding dynamic equipment to be identified has an operating fault.

[0083] Specifically, to reduce noise interference, eliminate sudden noise, and improve accuracy, the solution of the present invention uses probability statistics to determine whether a fault exists. Probability statistics has two meanings. One is sampling acoustic signals along a linear distance, for example, within a certain distance (e.g., 50 cm) before and after the active equipment to be identified, i.e., within a distance range (1 m). The evaluation result is deemed abnormal only if 95% of the signal identification results are abnormal. The other is based on the sensor (pickup) array. The acoustic signal collection probe can be single-channel, double-channel, or multi-channel array sensing. Single-channel probability statistics only apply to linear distance. In double-channel and multi-channel array sensing, the signals collected by each sensor and the signals collected at linear distance points form a matrix. The evaluation result is deemed abnormal only if 95% of the signal identification results are abnormal.

[0084] Step S40: Based on the feature information, the selected fault diagnosis model is trained, and the faulty equipment is identified based on the training result.

[0085] After the feature data is obtained, the feature data can be directly used as input parameters to train the correspondingly selected fault diagnosis model, and the fault diagnosis result can be obtained based on the model training result.

[0086] Preferably, in order to further ensure the accuracy of the fault identification results, the solution of the present invention further proposes a fault diagnosis result verification method, which includes: obtaining sound field distribution map information for each audio frequency band, comparing the sound field distribution map information for each frequency band with preset standard sound field distribution map information, obtaining the sound field distribution map deviation degree for each frequency band, obtaining a total deviation degree matrix based on the sound field distribution map deviation degrees for each frequency band, training a preset sound field fault identification model using the total deviation degree matrix as an input parameter, obtaining equipment abnormality results in the current operating state, comparing the equipment abnormality results obtained based on the sound field fault identification model with the equipment abnormality results obtained based on the fault diagnosis model, and verifying the equipment abnormality results obtained by the current fault diagnosis model.

[0087] Specifically, because sound propagation is overlapping, noise in complex on-site environments is often the result of the overlap of multiple sound sources, and changes in the spectral characteristics of the sound emitted by the sound source, changes in the spatial distribution of the sound source, and increases or decreases in the number of sound sources will all cause changes in the sound waveform collected at a fixed detection point. Based on this, the verification method proposed in this invention uses acoustic imaging technology to collect noise audio data and sound field video data of the object being measured, and through frequency band discrimination and judgment, comprehensively determine the sound spectral characteristics and distribution state within the entire frequency range, thereby achieving more comprehensive, accurate, and intuitive discrimination and visualization of various types of abnormal behavior.

[0088] Specifically, the fault identification results of the comparison configuration must first be obtained, and then the two fault identification results are compared. If the difference between the two meets the requirement, the fault identification result is deemed to meet the requirement. If the difference between the two exceeds the requirement, it indicates that a certain configuration has been identified incorrectly, and fault identification must be attempted again to ensure identification accuracy. Preferably, the position of the collection robot is adjusted and fault identification must be attempted again until both fault identification results meet the requirement.

[0089] When obtaining the comparison results, the corresponding acoustic field fault identification model must first be constructed. When the on-site equipment is in normal operating mode, the acoustic imaging equipment is used to collect and record video and audio signals, and a sample library of normal operating mode is established. The steps are as follows:

[0090] First, select the total frequency range [Φmin, Φmax] for sound analysis, where Φmax is less than half the microphone sampling frequency. Then, select N groups of characteristic frequency ranges based on the noise frequency characteristics of the target abnormal operating state. Divide the total frequency range [Φmin, Φmax] into N groups, with the frequency ranges of each group being [Φ1, Φ2], [Φ2, Φ3]...[ΦN-1, ΦN], where Φ1 = Φmin and ΦN = Φmax. Next, set the sound frequency range [Φ1, Φ2] for the first group and calculate the sound field distribution map. Record the coordinate γ1[x1, y1] of the maximum sound source point. Then, adjust the sound frequency ranges for the second, third, and Nth groups. Repeat the above steps to obtain the sound source point coordinates γ2[x2, y2], γ3[x3, y3]...γN[xN, yN] for each frequency band. The sound energy signal of a microphone in the microphone array is then calculated.

number

number

number

number

number

number

[0091] After obtaining the sample library of normal operating modes, in the case of abnormal operating modes, or when using a simulation experimental device to conduct a simulation experiment of abnormal operating states, the acoustic imaging equipment is used to collect and record video and audio signals, and the abnormal operating state identification model is trained. The steps are as follows:

[0092] Spectral density sequence in the first group of speech analysis frequency range [Φ1,Φ2]

number

number

number

number

number

number

number

number

[0093] The total deviation degree matrix is ​​taken as input, the type number of the abnormal operating state is taken as output, and the abnormal operating state identification model is established through training.

[0094] After obtaining the sound field fault identification model, the total deviation degree matrix is ​​used as an input parameter to train the preset sound field fault identification model, and obtain equipment abnormality results in the current operating state. The equipment abnormality results obtained based on the sound field fault identification model are compared with the equipment abnormality results obtained based on the fault diagnosis model, and the equipment abnormality results obtained by the current fault diagnosis model are verified.

[0095] Specifically, the entropy of the autocorrelation function of the detection signal of one segment is compared with a predetermined entropy threshold, and if the entropy of the autocorrelation function is equal to or less than the predetermined entropy threshold, the fault type of the equipment generating the detection signal of that segment is a mechanical fault type, and if the entropy of the autocorrelation function is greater than the predetermined entropy threshold, the fault type of the equipment generating the detection signal of that segment is a fluid dynamics anomaly type. When the current fault type diagnostic model is a mechanical fault diagnostic model, the step of extracting features from the detection signal and obtaining feature information includes a step of performing a fusion analysis of a singular spectrum and a higher-order spectrum on the detection signal, obtaining post-analysis data, and setting the analyzed data as feature information. When the current fault type diagnostic model is a fluid dynamics fault diagnostic model, the step of extracting features from the detection signal and obtaining feature information includes a step of performing a short-time Fourier transform on the detection signal, obtaining a corresponding spectral image, and setting the analyzed data as feature information.

[0096] Specifically, the step of training a selected fault diagnosis model based on the feature information and identifying faulty equipment based on the training result includes the step of using the feature information and operation process parameters as input parameters to train based on the machine fault diagnosis model and taking the acquired training result as a machine fault type analysis result, or the step of using the feature information and operation process parameters as input parameters to train the fluid mechanics fault diagnosis model and taking the acquired training result as a fluid mechanics anomaly type analysis result, and during the training process of the fluid mechanics anomaly identification model, the training process is interfered with based on an image data enlargement method.

[0097] Example 4

[0098] Within each audio frequency band, sound field distribution map information for each frequency band is obtained, the sound field distribution map information for each frequency band is compared with preset standard sound field distribution map information, the sound field distribution map deviation degree for each frequency band is obtained, a total deviation degree matrix is ​​obtained based on the sound field distribution map deviation degree for each frequency band, the total deviation degree matrix is ​​used as an input parameter to train a preset sound field fault identification model, obtain equipment abnormality results in the current operating state, compare the equipment abnormality results obtained based on the sound field fault identification model with the equipment abnormality results obtained based on the fault diagnosis model, and verify the equipment abnormality results obtained by the current fault diagnosis model.

[0099] Specifically, whether it is a mechanical fault type identification result or a fluid dynamics abnormality type identification result, the fault type information that generates the voice can be finally estimated based on the corresponding voice feature information, and the fault type information can be directly pushed to the user terminal to assist the user in identifying and investigating the corresponding fault.

[0100] Preferably, in order to further ensure the accuracy of the fault identification results, the solution of the present invention further proposes a fault diagnosis result verification method, which includes: obtaining sound field distribution map information for each audio frequency band, comparing the sound field distribution map information for each frequency band with preset standard sound field distribution map information, obtaining the sound field distribution map deviation degree for each frequency band, obtaining a total deviation degree matrix based on the sound field distribution map deviation degrees for each frequency band, training a preset sound field fault identification model using the total deviation degree matrix as an input parameter, obtaining equipment abnormality results in the current operating state, comparing the equipment abnormality results obtained based on the sound field fault identification model with the equipment abnormality results obtained based on the fault diagnosis model, and verifying the equipment abnormality results obtained by the current fault diagnosis model.

[0101] Specifically, because sound propagation is overlapping, noise in complex on-site environments is often the result of the overlap of multiple sound sources, and changes in the spectral characteristics of the sound emitted by the sound source, changes in the spatial distribution of the sound source, and increases or decreases in the number of sound sources will all cause changes in the sound waveform collected at a fixed detection point. Based on this, the verification method proposed in this invention uses acoustic imaging technology to collect noise audio data and sound field video data of the object being measured, and through frequency band discrimination and judgment, comprehensively determine the sound spectral characteristics and distribution state within the entire frequency range, thereby achieving more comprehensive, accurate, and intuitive discrimination and visualization of various types of abnormal behavior.

[0102] Specifically, the fault identification results of the comparison form must first be obtained, and then the two fault identification results are compared. If the difference between the two meets the requirement, it indicates that the fault identification result meets the requirement. If the difference between the two exceeds the requirement, it indicates that there is a recognition error in a certain form, and fault identification must be attempted again to ensure recognition accuracy.

[0103] When obtaining the comparative results, it is first necessary to build a corresponding acoustic field fault identification model. When the on-site equipment is in normal operation mode, acoustic imaging equipment is used to collect and record video and audio signals, and a sample library of normal operation mode is established.

[0104] After obtaining the sound field fault identification model, the total deviation degree matrix is ​​used as an input parameter to train the preset sound field fault identification model, and obtain the equipment abnormality result in the current operating state. The equipment abnormality result obtained based on the sound field fault identification model is compared with the equipment abnormality result obtained based on the fault diagnosis model, and the equipment abnormality result obtained by the current fault diagnosis model is verified. After the consistency test between the two is passed, the final fault diagnosis result is determined.

[0105] Example 5

[0106] Various fault types were artificially labeled, including bearing wear, misalignment, imbalance, machine looseness, shaft misalignment, loose foundation, cavitation, and turbulence. For the same rotating equipment, only one fault type was labeled at a time, and 200 groups of data were collected for each fault type, divided by time and background. Finally, each fault type had 200 groups of data, for a total of 1,600 groups of data, each containing audio and vibration signals. Furthermore, the 1,600 groups of data included data collected from multiple pre-defined points, each at a different distance and orientation from the rotating equipment. These data were used as comparative samples to form multiple comparative examples.

[0107] Comparative Example 1: The fault type of each group of data is hidden, and all data are randomly combined to obtain a sequence of data groups with unknown fault types and confusing sequence numbers. Based on a fault detection algorithm combined with a conventional fixed fault sensor, corresponding fault detection is performed to obtain the diagnostic fault type of each group of data. The hidden fault types of the corresponding data are then displayed and compared, and the data groups with the same fault type are 1432 groups.

[0108] Comparative Example 2: The fault type of each group of data is hidden, and all data is randomly combined to obtain a sequence of data groups with unknown fault types and confusing sequence numbers. Fault detection is performed based on a fusion fault diagnosis model based on acoustic and vibration signals acquired through simulation, and the diagnosed fault type of each group of data is obtained. Then, the hidden fault types of the corresponding data are displayed and compared. The data groups with the same fault types are 1553 groups.

[0109] Comparative Example 3: The fault type of each group of data is hidden, and all data is randomly combined to obtain a sequence of data groups with unknown fault types and confusing sequence numbers. Fault detection is performed based on a fusion fault diagnosis model based on acoustic and vibration signals obtained through simulation, and the diagnosed fault type of each group of data is obtained. The detection results are corrected based on the deviation degree of the sound field distribution map in each frequency band. The corrected detection results are compared with the original results, and there are 1572 data groups where the two are the same.

[0110] It can be seen that the fault identification accuracy of the fusion fault diagnosis model based on acoustic signals and vibration signals according to the solution of the present invention is significantly higher than that of the conventional fault algorithm based on a single signal.

[0111] 4 is a diagram illustrating the configuration of a mobile fault detection device for dynamic equipment according to an embodiment of the present invention. As shown in FIG. 4, an embodiment of the present invention provides a mobile fault detection device for dynamic equipment, which includes a signal collection robot, a processing unit, and a training unit.

[0112] The signal collection robot is used to patrol and collect detection signals within the equipment installation area, the processing unit is used to classify the detection signals, select a pre-set fault diagnosis model based on the classification result, and perform feature extraction on the detection signals to obtain feature information, and the training unit is used to train the selected fault diagnosis model based on the feature information and identify faulty equipment based on the training result.

[0113] Optionally, the signal collection robot includes a collection module, a traveling module, and a navigation module, wherein the collection module includes a plurality of acoustic signal collectors arranged in a circular array and used for synchronously collecting acoustic signals generated from the target dynamic equipment, and a laser vibrometer used for emitting a vibration measurement laser to the target dynamic equipment and collecting corresponding vibration signals, wherein the traveling module is used for driving the patrol inspection and movement of the patrol inspection robot, and the navigation module is used for obstacle avoidance during the movement of the patrol inspection robot.

[0114] A third aspect of the present invention provides a mobile fault detection system, said system including the mobile fault detection facility for the dynamic equipment described above.

[0115] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, the instructions, when executed by a computer, causing the computer to perform the mobile fault detection method for dynamic equipment described above.

[0116] As can be understood by those skilled in the art, all or part of the steps in the methods of the above embodiments can be achieved by instructing relevant hardware through a program, which is stored in a storage medium and includes a plurality of instructions for causing a one-chip microcomputer, chip, or processor to perform all or part of the steps of the methods of the above embodiments. The storage medium includes various media capable of storing program code, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0117] Although the above describes in detail the optional embodiments of the present invention with reference to the drawings, the embodiments of the present invention are not limited to the details in the above embodiments. Various simple modifications can be made to the technical solutions of the embodiments of the present invention within the technical concept of the embodiments of the present invention, and all of these simple modifications fall within the scope of protection of the embodiments of the present invention. It should be noted that the specific technical features described in the above specific embodiments can be combined in any appropriate manner if not contradictory. In order to avoid unnecessary repetition, the embodiments of the present invention will not separately describe various possible combinations.

[0118] Furthermore, various different embodiments of the present invention can be combined in any manner, and as long as the combination does not deviate from the technical idea of ​​the embodiments of the present invention, the combination should also be considered to be the content disclosed in the embodiments of the present invention.

Claims

1. 1. A mobile fault detection method for dynamic equipment, the method comprising: A step of circulating and collecting detection signals within an equipment installation area; classifying the detection signal and selecting a preset fault diagnosis model based on the classification result; performing feature extraction on the detection signal to obtain feature information; training a selected fault diagnosis model based on the feature information, and identifying faulty equipment based on the training results.

2. The detection signal is acoustic and / or vibration signals generated in the target facility; and location information of the target equipment at the current collection time and an operating process parameter of the target equipment.

3. The operating process parameters are:

3. The method of claim 2, including one or more of operating temperature information, pressure information, flow rate information, and fixation method information.

4. The step of cyclically collecting detection signals within an equipment installation area includes: determining a filtering limit margin of the target equipment based on the location information of the current collection time and the operating parameters of the target equipment; determining an actual filtering limit value of the target equipment based on the filtering limit value margin and a preset basic filtering limit value; and collecting detection signals of the target facility based on the actual filtering limit values.

5. The step of determining a filtering limit margin of the target equipment based on the location information of the current collection time and the operating parameters of the target equipment includes: determining distance information between the location at the current collection time and the target facility based on the location information at the current collection time; determining detection signal prediction information for the target equipment based on the operating parameters of the target equipment; determining a filtering limit margin for the target facility based on the distance information and the detection signal prediction information; The detection signal prediction information is 5. The method of claim 4, including predicting one or more of the amplitude, frequency, and wavelength of the detected signal.

6. The step of determining a filtering limit margin of the target equipment based on the distance information and the detection signal prediction information includes: determining an attenuation rule of the detection signal prediction information based on the distance information; determining detection signal prediction information for the position of the current collection time based on the attenuation rule; and determining a filtering threshold margin for the target facility based on the detected signal prediction information for the current collection time position.

7. The step of classifying and processing the detection signal includes: determining a relationship between the currently collected acoustic signal and the vibration signal and a preset minimum acoustic signal strength and a preset vibration signal strength, respectively, and classifying the detected signal based on the determination result; If the currently collected acoustic signal is smaller than the predetermined minimum acoustic signal strength but the vibration signal is equal to or greater than the predetermined vibration signal strength, the classification result of the currently detected signal is a vibration signal; If the currently collected acoustic signal is equal to or greater than the predetermined minimum acoustic signal strength but the vibration signal is smaller than the predetermined vibration signal strength, classifying the currently detected signal as an acoustic signal; If the currently collected acoustic signal is equal to or greater than the predetermined minimum acoustic signal strength and the vibration signal is equal to or greater than the predetermined vibration signal strength, the classification result of the currently detected signal is a fusion signal of the vibration signal and the acoustic signal; 3. The method according to claim 2, further comprising the step of: if the currently collected acoustic signal is smaller than the preset minimum acoustic signal strength and the vibration signal is smaller than the preset vibration signal strength, the currently collected signal does not meet the requirements, and it is necessary to adjust the signal collection position and collect the detection signal again.

8. The preset fault diagnosis model is The method of claim 7 , comprising a fault diagnosis model based on acoustic signals, a fault diagnosis model based on vibration signals, and a fusion fault diagnosis model based on acoustic and vibration signals.

9. The step of selecting a preset fault diagnosis model based on the classification result includes: If the classification result is a vibration signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on a vibration signal; If the classification result is an acoustic signal, the correspondingly selected preset fault diagnosis model is a fault diagnosis model based on an acoustic signal; and if the classification result is a fusion signal, the correspondingly selected preset fault diagnosis model is a fusion fault diagnosis model based on an acoustic signal and a vibration signal.

10. When the current model is a fusion fault diagnosis model based on an acoustic signal and a vibration signal, the step of extracting features from the detection signal and obtaining feature information includes: performing feature extraction and feature fusion on the detection signal based on a preset EMD decomposition algorithm to obtain a fusion feature vector set; 9. The method of claim 8, further comprising: performing fault word replacement on each fused feature vector of the fused feature vector set based on a preset fused bag-of-words model to obtain a word evaluation set as feature information.

11. The step of performing feature extraction on the detection signal based on a preset EMD decomposition algorithm to perform feature fusion and obtain a fusion feature vector set includes: performing EMD decomposition on the synchronously collected acoustic and vibration signals to obtain multiple signal components for each signal; performing an HHT transform on each signal component to obtain an instantaneous frequency and an instantaneous amplitude of each signal component; 11. The method of claim 10, further comprising: comparing the instantaneous frequencies of the signal components of the acoustic signal with the instantaneous frequencies of the signal components of the vibration signal; adding the instantaneous amplitudes corresponding to the same points of the instantaneous frequencies of both signals to obtain a plurality of fused signals; and constructing and obtaining corresponding time-frequency matrices as a fused feature vector set.

12. The pre-configured fused bag-of-words model is a fault word substitution model used to perform fault word substitution based on the fused feature vector; and a predetermined codebook used to record the correspondence between the word frequency vector and the fault cause.

13. The method comprises: Further comprising the step of constructing a faulty word replacement model, which step comprises: A step of collecting detection signals in a fault state in historical data of the dynamic equipment, or collecting detection signals in an operating state of the dynamic equipment labeled as being in a fault state, and using them as basic data; performing feature extraction on the basic data and performing feature fusion to obtain a fusion feature vector training set; 11. The method of claim 10, further comprising: clustering the fused feature vector training set based on a preset clustering algorithm to obtain a plurality of clustering centers; performing fault word labeling for each clustering center; and obtaining a fault word substitution model.

14. The step of training a selected fault diagnosis model based on the feature information and identifying faulty equipment based on the training result includes: introducing the word evaluation set and the operation process parameters of the target equipment as input parameters into the fusion fault diagnosis model based on the acoustic signal and the vibration signal; A step of performing statistics on the occurrence frequency of fault words based on a preset clustering algorithm; Obtaining a corresponding word frequency vector according to the statistical result of the occurrence frequency of the fault word; and identifying the cause of the fault based on the word frequency vector against a pre-established codebook.

15. The fault diagnosis model further includes a faulty equipment location identification model; The step of identifying faulty equipment based on the training results includes: The method further includes a step of identifying a faulty equipment location, the step including: a step of cyclically collecting acoustic signals to be identified within a predetermined collection area, recording collection position information for each collection time of the acoustic signals to be identified, and obtaining a correspondence relationship between a plurality of groups of acoustic signals to be identified and the collection position information; 3. The method according to claim 2, further comprising: training a faulty equipment location identification model using all acoustic signals to be identified and the corresponding collected location information as input parameters, and obtaining location information of the faulty equipment.

16. The step of circularly collecting acoustic signals to be identified within a predetermined collection area includes: The method includes a step of collecting a flow of acoustic signals to be identified by linearly traveling along any two opposite directions starting from the dynamic equipment until the method reaches a position having a predetermined maximum distance from the dynamic equipment; The acquisition position of the acoustic signal to be identified is different each time, and the distance between the acquisition positions of any two adjacent acoustic signals to be identified is the same; 16. The method of claim 15, wherein at each acoustic signal acquisition location, a plurality of acoustic signals to be differentiated are acquired synchronously in a multi-channel manner.

17. The step of recording collection position information for each collection time of the acoustic signal to be identified includes: Responding to a collection trigger signal of the acoustic signal, reading laser navigation data circulating at a current collection position; reading first collection position candidate information of a current collection position based on the laser navigation data; reading Beidou positioning information of a current collection position in response to a collection trigger signal of an acoustic signal; reading second collection position candidate information of the current collection position based on the Beidou positioning information; The method according to claim 15, further comprising: correcting the collection position based on the first collection position information and the second collection position information to obtain the collection position.

18. The method comprises: In each audio frequency band, obtain sound field distribution map information of each frequency band, compare the sound field distribution map information of each frequency band with preset standard sound field distribution map information, and obtain the sound field distribution map deviation degree of each frequency band; obtaining a total deviation degree matrix based on the sound field distribution map deviation degree of each frequency band; Using the total deviation degree matrix as an input parameter, training a pre-defined sound field fault identification model to obtain equipment abnormality results in the current operating state; 2. The method according to claim 1, further comprising: comparing the equipment abnormality results obtained based on the sound field fault identification model with the equipment abnormality results obtained based on the fault diagnosis model, and verifying the equipment abnormality results obtained using the current fault diagnosis model.

19. A mobile fault detection facility for dynamic installations, the facility including a signal collection robot, a processing unit, and a training unit; The signal collecting robot is used to patrol and collect detection signals within the equipment installation area; The processing unit classifying the detection signal and selecting a preset fault diagnosis model based on the classification result; extracting features from the detection signal to obtain feature information; The mobile fault detection equipment for dynamic equipment is characterized in that the training unit is used to train a selected fault diagnosis model based on the feature information, and to identify faulty equipment based on the training result.

20. The detection signal is acoustic and / or vibration signals generated in the target facility; 20. The facility of claim 19, further comprising: location information of the current collection time and operating process parameters of the target facility.

21. The signal collecting robot includes a collecting module, a traveling module, and a navigation module; The collection module includes: a plurality of acoustic signal collectors arranged in a circular array and used for synchronously collecting acoustic signals generated by the target dynamic equipment; a laser vibrometer used for emitting a vibration measurement laser to a target dynamic equipment and collecting corresponding vibration signals; The traveling module is used to drive the patrol inspection and movement of the patrol inspection robot; The facility according to claim 19, wherein the navigation module is used for obstacle avoidance during the movement of the patrol inspection robot.

22. A mobile fault detection system for dynamic equipment, comprising the mobile fault detection equipment for dynamic equipment according to any one of claims 19 to 21.

23. A computer-readable storage medium having stored thereon instructions that, when executed by a computer, cause the computer to perform the method for mobile fault detection in dynamic equipment according to any one of claims 1 to 18.

Images