Systems and methods for audio and vibration based condition monitoring of power distribution equipment
Through an intelligent analysis system based on audio and vibration, combined with machine learning algorithms, the operating status of the power distribution equipment is monitored in real time, and the problem of difficult to monitor and predict equipment abnormalities in the prior art is solved, and the effect of reducing maintenance costs and downtime is achieved.
Patent Information
- Application Number
- CN201980015851.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2018-02-28
- Filing Date
- 2019-02-26
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2039-02-26
AI Technical Summary
Existing equipment health monitoring systems are difficult to effectively monitor and predict abnormal situations of power distribution equipment, resulting in high maintenance costs and long system downtime.
Using an intelligent analysis system based on audio and vibration, through IAA devices and sensing components, combined with machine learning algorithms such as deep convolutional neural networks and deep recursive neural networks, the operating status of power distribution devices is monitored and predicted in real time to identify abnormal situations. .
Real-time abnormality detection and prediction of power distribution equipment is realized, which reduces maintenance costs, reduces system downtime, and improves equipment reliability and efficiency.
Smart Images

Figure CN111742462B_ABST
Abstract
Description
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims priority to U.S. Provisional Application Serial No. 61 / 715,038, filed by Das et al. on February 28, 2018, entitled “SYSTEM AND METHOD FOR AUDIO AND VIBRATION BASED POWER DISTRIBUTION EQUIPMENT CONDITION MONITORING,” the disclosure of which is hereby incorporated by reference in its entirety. Technical Field
[0003] The present disclosure relates generally to equipment health monitoring systems, and more particularly to audio and vibration based power distribution equipment condition monitoring systems and methods. Background Art
[0004] Unless otherwise indicated herein, the materials described in this section are not prior art to the claims in this application and are not admitted to be prior art by inclusion in this section. Summary of the invention
[0005] The following describes an overview of some embodiments disclosed herein. It should be understood that these aspects are presented only to provide the reader with a brief overview of these embodiments, and these aspects are not intended to limit the scope of the present disclosure. In fact, the present disclosure may encompass a variety of aspects that may not be described below.
[0006] Embodiments of the present disclosure relate to a system and method for monitoring power distribution equipment condition based on audio and vibration. The system includes an IAA device and a sensing assembly. The IAA device having computer executable instructions (such as an audio data processing algorithm) is configured to identify and predict impending anomalies associated with one or more power distribution equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] These and other features, aspects and advantages of the present disclosure will become better understood when the following detailed description of certain exemplary embodiments is read with reference to the accompanying drawings, in which like characters represent like techniques throughout the drawings, and in which:
[0008] Figure 1 is a simplified block diagram of an audio and vibration based power distribution equipment condition monitoring system in accordance with described embodiments of the present disclosure;
[0009] Figure 2 is a simplified block diagram of an intelligent audio and vibration analysis system according to a described embodiment of the present disclosure;
[0010] Figure 3is a simplified block diagram of an intelligent processor module according to a described embodiment of the present disclosure;
[0011] Figure 4 is a simplified autoencoder network architecture according to the described embodiments of the present disclosure;
[0012] Figure 5 is a diagram of a deep recurrent neural network (DRNN) that receives a vector of audio states according to described embodiments of the present disclosure; and
[0013] Figure 6 is a diagram of a deep convolutional neural network (CNN) for continuously monitoring machine operating status according to a described embodiment of the present disclosure. DETAILED DESCRIPTION
[0014] For the purpose of promoting an understanding of the principles of the present disclosure, reference will now be made to the embodiments illustrated in the accompanying drawings and described in the following written specification. It should be understood that no limitation of the scope of the present disclosure is thereby intended. It should be further understood that the present disclosure includes any changes and modifications to the illustrated embodiments, and includes further applications of the principles of the present disclosure as would normally occur to one of ordinary skill in the art to which the present disclosure belongs.
[0015] Various operations may be described in turn as multiple discrete actions or operations in a manner that is most helpful for understanding the claimed subject matter. However, the order of description should not be interpreted as implying that these operations must rely on order. In particular, these operations may not be performed in the order presented. The described operations may be performed in an order different from the described embodiments. In additional embodiments, various additional operations may be performed and / or the described operations may be omitted.
[0016] Figure 1is a simplified block diagram of an audio and vibration based power distribution equipment condition monitoring system 110 in an exemplary network environment 100 according to a described embodiment of the present disclosure. The system 110 includes an intelligent audio analysis (IAA) device 112, which includes an audio data processing algorithm programmed into the IAA device 112 and is configured to identify and predict impending anomalies associated with any suitable power distribution equipment 120. The power distribution equipment 120 may include transformers, regulators, chillers, reclosers, etc. Depending on the size of the facility, any suitable number of power distribution equipment 120 may be installed in the exemplary network environment. The functions performed by the IAA device 112 allow for classification of both erroneous machine operation and deteriorating machine health or condition, thereby facilitating preventive or corrective actions, thereby reducing maintenance costs and limiting system downtime. The IAA device module 112 analyzes the temporal dynamics in the audio / vibration data and utilizes state-of-the-art machine learning algorithms for time series anomaly detection and prediction.
[0017] Figure 2 1 is a simplified block diagram of an intelligent audio and vibration analysis system 112 or “IAA device module” as defined herein, in accordance with a described embodiment of the present disclosure. The IAA device module 112 may be incorporated into the power distribution equipment monitoring system 110, or may be a separate module independent of the power distribution equipment monitoring system and communicatively coupled to the power distribution equipment monitoring system 110. Figure 2In one embodiment, the IAA device 112 is communicatively coupled to the sensing assembly 130. In one embodiment, the sensing assembly may include a vibration sensor, an accelerometer, an audio input, a motion sensor, or any suitable sensing unit. As illustrated, an accelerometer 130a and a microphone 130b are included in the sensing assembly 130. In order to collect training data for the machine learning algorithm, the sensing assembly is mounted on or near the power distribution device under test (DUT). The sensing assembly 120 detects the signal, and then the detected signal is preprocessed to remove any competitive environment or surrounding noise, such as passing vehicles, road works, etc. In one embodiment, the preprocessing function may be performed by the sensing assembly 120. In some embodiments, the preprocessing function may be performed by a preprocessor, which is disposed in the sensing assembly 130 to form part of the sensing assembly, or is located outside the sensing assembly 130 and is communicatively coupled to the sensing assembly 130. If the preprocessor is located outside the sensing assembly 130, the preprocessor may be carried by the IAA device 112 to form part of the IAA device 112. Alternatively, the preprocessor can be an independent computer module. The sensing assembly 130 can not only use the audio input 130b to capture audio signals, but the sensing assembly 130 can also use the vibration sensor to capture vibration signals. Once the audio and / or vibration signals are captured, the IAA device 112 estimates the operating state of the power distribution equipment 120 to determine whether the equipment 120 is normal or abnormal. The ability to monitor / track the operating state of a machine (such as a distribution transformer) reveals how an individual machine reacts to the introduction of different types of loads (e.g., a solar array). In the IAA device 112, a processor 112a, a memory 112b, and a communication module 112c are provided. Depending on the requirements of the application, other computer modules (such as, a user interface) or any suitable module can be included in the IAA device 112. An audio analysis algorithm is installed in the processor 112a for execution. A further detailed description of the audio analysis algorithm will be provided below. Processor 112 receives classification parameters for audio-based machine state estimation from memory 112b, and receives real-time audio and / or vibration streams from equipment 120 via sensing assembly 130, and outputs inferred machine states, such as specific impacts on the operating state of equipment 120 (such as a transformer) due to specific loads (such as solar array integration). The estimated machine state or anomaly is then passed to communication module 112c, which in turn transmits to external devices for further statistical analysis.
[0018] Once the impact of load distribution / characteristics is tracked for individual machines 120, an external device (such as a cloud computing module 140) communicatively coupled to the IAA device 112 can further perform a large-scale spatiotemporal analysis of the impact of load distribution on the entire power distribution equipment condition monitoring system 110 by utilizing a machine condition monitoring algorithm based on audio analysis to aggregate results from various locations (including each instance and or location). In some embodiments, the cloud computing module can be located in a cloud network. In another embodiment, the cloud computing module is also a cloud network. Visualization tools such as programs or apps (applications) can be installed on a computer machine with a display to display the status of the ongoing power distribution process or the status of the entire power distribution infrastructure to the power distribution facility manager. The computer machine communicatively coupled to the cloud computing module can be a desktop computer, a portable computer, a laptop computer, a tablet computer, a wearable device, or any suitable computer machine. In the event that the IAA device 112 detects an abnormal machine state in the power distribution equipment 120, a notification system 140a (such as a tool, program, or app) is used to notify appropriate personnel (such as facility managers). The notification system 140a can run on any suitable computer machine. The notification may be in the form of a text message, email, voicemail, audible sound, display, indicator, or any suitable notification form, depending on the requirements of the application. In some embodiments, any of the notification system 140a, visualization tool, and cloud computing module 140b may run on a single computer machine, multiple machines, a network, a server, or a combination thereof. In some embodiments, the notification system 140a, visualization tool, and cloud computing module 140b may be installed at the same site, different sites, or multiple sites.
[0019] Figure 3 is a simplified block diagram of an intelligent processor module 112a according to a described embodiment of the present disclosure. In one embodiment, the intelligent processor module 112a is arranged Figure 2In some embodiments, the intelligent processor module 112a can be a separate and independent module and is communicatively coupled to the sensing assembly 130 and the IAA device 112. The audio signal processing installs the audio analysis algorithm and provides power for the audio analysis algorithm for audio feature representation and machine learning capabilities. The signal processing and machine learning algorithms behind the aforementioned audio analysis capabilities are largely agnostic to the specific sensing modality and can therefore be adapted to any of the available time series signals (e.g., gyroscopes, magnetometers, current, impedance, and capacitance) depending on the application scenario with the power distribution system. In fact, expanding the algorithm to incorporate multimodal sensory fusion will enable the use of deeper, possibly hidden dependencies and correlations, and thereby improve the prediction and detection of anomalies. As depicted in the figure, the processor 112a includes several computer executable programs or functions, such as scene grammars, pattern classification learners, supervisory signals, and audio feature representations. The processor 112a may also include deep learning, audio feature learning, and recognition features or programs. Other computer executable programs or instructions may be installed in and executed by the processor 112a depending on the requirements of the application.
[0020] Figure 4 is a simplified autoencoder network architecture according to the described embodiment of the present disclosure. For anomaly detection, an autoencoder (AE) and several of its derivatives, such as denoising AE, deep AE, to name a few examples, can be used. AE relies on a low-dimensional latent representation of a signal subspace that can distinguish between normal and abnormal data when reconstructed and coupled with an error threshold, i.e., anomalies tend to have larger reconstruction errors. The aforementioned derivatives of AE improve the performance of standard AE by adding robustness and exploiting deeper structures that may exist in the data. As depicted in the figure, the first layer L1 corresponds to the input signal (e.g., the FFT of an audio / vibration signal from a power distribution machine), and the third layer L3 is a reconstructed version of the signal. The second layer L2 is a low-dimensional latent representation. The encoder can be trained with machine data, assuming that abnormal operations are sparse over time. For those outlier signals, the reconstruction error at the third layer L3 will be large. The error magnitude can be used as a hint for anomaly detection.
[0021] Figure 5is a diagram of a deep recurrent neural network (DRNN) that receives a vector of audio states according to a described embodiment of the present disclosure. Unlike AE, only the frame or instantaneous difference between the current data and the known classifier / anomaly is utilized. In addition, AE does not retain any temporal information that can be used to predict the underlying structure that caused the anomaly. In order to fully utilize temporal knowledge to assist as preventive maintenance, a deep recurrent neural network (DRNN) is one of the methods based on deep learning to predict when possible anomalies will occur.
[0022] Compared to frame-based approaches, DRNNs are able to model the temporal dynamics of audio patterns, in which each frame is associated with a label (e.g., normal vs. abnormal) without taking temporal context / variations into account. However, training DRNNs on machine operation time series data involves pre-annotating the collected audio patterns into different operating modes. Using that annotated data, DRNNs can learn significant audio patterns associated with normal power distribution operations, as well as detect when they deviate into anomalies by identifying fault modes (e.g., failure of a particular transformer unit). The DRNN executed by the power distribution equipment condition monitoring system 110 not only demonstrates an excellent ability to model audio signals associated with human speech, but the power distribution equipment condition monitoring system 110 is also able to perform monitoring and anomaly detection / prediction.
[0023] Figure 6 is a diagram of a deep convolutional neural network (CNN) for continuously monitoring machine operating states according to a described embodiment of the present disclosure. The deep CNN for continuously monitoring machine operating states can be installed on the power distribution equipment condition monitoring system 110 to perform anomaly detection. The deep CNN further facilitates statistical analysis in the cloud network for data mining and optimizing machine operations, such as the long-term impact of loads (such as solar arrays) integrated into the power grid. CNN-based analysis is applicable to audio / vibration associated with all power distribution equipment (such as transformers, regulators, and reclosers). In some embodiments, CNN can be used in a supervised learning framework. For example, first, time-stamped data can be collected for different machine states, and the corresponding time audio-vibration segments can be transformed into spectrograms for further analysis. Thereafter, the spectrogram is treated as a natural 2D image and fed through a VGG network (a very deep CNN designed for image recognition) to identify specific machine states. An overview of the deep CNN architecture performed by the power distribution equipment condition monitoring system 110 is provided in the figure.
[0024] The embodiments described above have been shown by way of example, and it should be understood that these embodiments may be susceptible to various modifications and alternative forms. It should be further understood that the claims are not intended to be limited to the specific forms disclosed, but to cover all modifications, equivalents and alternatives that fall within the spirit and scope of the present disclosure.
[0025] Embodiments within the scope of the present disclosure may also include non-transitory computer-readable storage media or machine-readable media for carrying or having computer-executable instructions or data structures stored thereon. Such non-transitory computer-readable storage media or machine-readable media may be any available media that can be accessed by a general or special-purpose computer. By way of example and not limitation, such non-transitory computer-readable storage media or machine-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, or any other medium that may be used to carry or store desired program code components in the form of computer-executable instructions or data structures. The above combination should also be included in the scope of non-transitory computer-readable storage media or machine-readable media.
[0026] Embodiments may also be practiced in distributed computing environments where tasks are performed by local and remote processing devices that are linked (either by hardwired links, wireless links, or by a combination thereof) through a communications network.
[0027] Computer executable instructions include, for example, instructions and data that cause a general-purpose computer, a special-purpose computer, or a special-purpose processing device to perform a certain function or group of functions. Computer executable instructions also include program modules that are executed by a stand-alone computer or a computer in a network environment. In general, program modules include routines, programs, objects, components, and data structures that perform specific tasks or implement specific abstract data types. Computer executable instructions, associated data structures, and program modules represent examples of program code components for performing the steps of the method disclosed herein. A specific order of such executable instructions or associated data structures represents an example of corresponding actions for implementing the functions described in such steps.
[0028] Although the present disclosure has been illustrated and described in detail in the drawings and the foregoing description, it should be considered illustrative rather than restrictive in nature. It should be understood that only preferred embodiments have been presented and all changes, modifications and further applications falling within the spirit of the present disclosure are desired to be protected.
Claims
1. A power distribution equipment monitoring system, comprising: a plurality of sensing assemblies, each sensing assembly configured for incorporation into, onto, or proximate to a respective power distribution equipment of a plurality of power distribution equipment, each of the plurality of sensing assemblies configured to output a respective sensor signal indicative of sound and / or vibration from the respective power distribution equipment; an intelligent audio analysis device configured to receive sensor signals from the plurality of sensing assemblies, process the sensor signals to detect anomalies from the operation of the corresponding power distribution equipment, and output detection information to an output device to alert an operator to the detection of any anomalies, wherein the intelligent audio analysis device includes programming instructions stored on a non-transitory computer-readable storage medium, and a processor configured to execute the programming instructions, wherein the programming instructions include an analysis algorithm for execution by a processor, the analysis algorithm causing the processor to continuously process the sensor signals to detect anomalies, wherein the analysis algorithm comprises at least one neural network, and wherein input to the at least one neural network is based on sensor signals received by an intelligent audio analysis device, Wherein once the impact of load distribution is tracked for individual power distribution equipment, the processor is configured to perform a spatiotemporal analysis of the impact of load distribution on the entire power distribution equipment monitoring system by aggregating results from the various locations.
2. The system of claim 1, wherein the sensor signal is pre-processed before being supplied as input to the neural network.
3. The system of claim 2, wherein the at least one neural network comprises an autoencoder neural network configured to receive a sensor signal as an input and output a reconstructed sensor signal based on the input, wherein an error signal is generated, the error signal being indicative of a difference between an input and an output of the autoencoder neural network, and Wherein the processor is configured to detect an anomaly based on a magnitude of the error signal. 4 . The system of claim 3 , wherein the processor is configured to compare the magnitude of the error signal with an error threshold, and detect an anomaly when the magnitude of the error signal exceeds the error threshold.
5. The system of claim 2, wherein the at least one neural network comprises a deep recurrent neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
6. The system of claim 2, wherein the at least one neural network comprises a deep convolutional neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
7. The system of claim 6, wherein the sensor signal is transformed into a spectrogram during preprocessing, the spectrogram being fed as input to a deep convolutional neural network, and Wherein the deep convolutional neural network is configured to process the spectrogram to identify anomalies.
8. A method for monitoring power distribution equipment, comprising: generating respective sensor signals indicative of sounds and / or vibrations originating from respective ones of a plurality of power distribution equipment, the respective sensor signals being detected by a plurality of sensor assemblies located on, in, or near the plurality of power distribution equipment; processing the respective sensor signals to detect anomalies from the operation of the respective power distribution equipment using a processor of the intelligent audio analysis device, the processor being configured to process the respective sensor signals to detect anomalies according to an analysis algorithm, the analysis algorithm comprising at least one neural network configured to detect errors and / or patterns indicative of anomalies in the sensor signals; and When an abnormality is detected, the detection information is output to an output device; Wherein once the impact of load distribution is tracked for individual power distribution equipment, the processor is configured to perform a spatiotemporal analysis of the impact of load distribution on the entire power distribution equipment monitoring system by aggregating results from the various locations.
9. The method of claim 8, wherein the sensor signal is pre-processed before being supplied as input to the neural network.
10. The method of claim 9, wherein the at least one neural network comprises an autoencoder neural network configured to receive a sensor signal as an input and to output a reconstructed sensor signal based on the input, wherein an error signal is generated, the error signal being indicative of a difference between an input and an output of the autoencoder neural network, and Wherein the processor is configured to detect an anomaly based on a magnitude of the error signal. 11 . The method of claim 10 , wherein the processor is configured to compare the magnitude of the error signal with an error threshold, and detect an anomaly when the magnitude of the error signal exceeds the error threshold.
12. The method of claim 9, wherein the at least one neural network comprises a deep recurrent neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
13. The method of claim 9, wherein the at least one neural network comprises a deep convolutional neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
14. The method of claim 13, wherein the sensor signal is transformed into a spectrogram during preprocessing, the spectrogram being fed as input to a deep convolutional neural network, and Wherein the deep convolutional neural network is configured to process the spectrogram to identify anomalies.
15. A power distribution equipment system, comprising: Multiple power distribution equipment; a plurality of sensing assemblies, each sensing assembly configured for incorporation into, onto, or proximate to a respective one of the plurality of power distribution equipment, the sensing assembly configured to output a sensor signal indicative of sound and / or vibration from the respective power distribution equipment; an intelligent audio analysis device configured to receive sensor signals from the plurality of sensing assemblies, process the sensor signals to detect anomalies from the operation of the corresponding power distribution equipment, and output detection information to an output device to alert an operator of the system of the detection of any anomalies, wherein the intelligent audio analysis device includes programming instructions stored on a non-transitory computer-readable storage medium, and a processor configured to execute the programming instructions, wherein the programming instructions include an analysis algorithm for execution by a processor, the analysis algorithm causing the processor to continuously process the sensor signals to detect anomalies, wherein the analysis algorithm comprises at least one neural network, and wherein input to the at least one neural network is based on sensor signals received by an intelligent audio analysis device, Wherein once the impact of load distribution is tracked for individual power distribution equipment, the processor is configured to perform a spatiotemporal analysis of the impact of load distribution on the entire power distribution equipment monitoring system by aggregating results from the various locations.
16. The system of claim 15, wherein the at least one neural network comprises an autoencoder neural network configured to receive a sensor signal as an input and output a reconstructed sensor signal based on the input, wherein an error signal is generated, the error signal being indicative of a difference between an input and an output of the autoencoder neural network, and Wherein the processor is configured to detect an anomaly based on a magnitude of the error signal. 17 . The system of claim 16 , wherein the processor is configured to compare the magnitude of the error signal to an error threshold, and detect an anomaly when the magnitude of the error signal exceeds the error threshold.
18. The system of claim 15, wherein the at least one neural network comprises a deep recurrent neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
19. The system of claim 15, wherein the at least one neural network comprises a deep convolutional neural network configured to receive sensor signals as input and detect audio and / or vibration patterns in the sensor signals that are indicative of anomalies.
20. The system of claim 15, wherein the sensor signal is transformed into a spectrogram during preprocessing, the spectrogram being fed as input to a deep convolutional neural network, and Wherein the deep convolutional neural network is configured to process the spectrogram to identify anomalies.
