Adaptive dual-entropy-based acoustic-vibration multi-modal edge cloud collaborative monitoring method

Abstract

This invention relates to the field of intelligent monitoring and proposes an acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy. By extracting features from multimodal signal sample data, it achieves second-level output of window-level state features and triggers alarms, meeting the real-time monitoring requirements for equipment start / stop identification and rapid response to abnormal operating conditions. It uses three-dimensional features to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices, significantly reducing the pressure on wireless link bandwidth and cloud storage and computing load. It can still connect a large number of devices and achieve long-term online monitoring in narrowband IoT environments, significantly improving scalability and cost controllability. It ensures real-time performance and reliability when deployed on a large scale in resource-constrained industrial sites. Based on multimodal feature fusion and discrimination, it can stably identify multi-level states, improving robustness and discrimination accuracy in complex operating conditions and high-noise environments. This invention improves the stability and accuracy of intelligent monitoring of electromechanical equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring, and in particular to a method for collaborative monitoring of acoustic-vibration multimodal edge clouds based on adaptive dual entropy. Background Technology

[0002] Currently, equipment condition monitoring technology based on the Industrial Internet of Things (IIoT) has been widely used in the industrial field.

[0003] Existing technologies mainly fall into two categories: the first is a device status monitoring system based on centralized cloud processing; the second is a cloud-edge collaborative fault diagnosis system that initially incorporates edge computing. For the first type of solution, its typical characteristic is that computationally intensive tasks are entirely placed in the cloud. After the wireless sensor collects the raw data, it is processed almost entirely offline, i.e., uploaded via LoRa or other links, with the cloud server completing core analysis tasks such as feature extraction and status assessment. Regarding the second type of solution, although the feature extraction task is offloaded to the edge, the signal processing and feature extraction algorithms it employs (such as Empirical Mode Decomposition (EMD)) have high computational complexity and consume significant resources. Therefore, existing technologies suffer from the following problems: First, high data transmission latency and poor real-time performance. The raw data, especially high-frequency sampled sound waves and vibration signals, is massive in volume, posing significant challenges in low-speed, high-concurrency NB-IoT or LoRa networks. The transmission in the network introduces significant transmission delays, resulting in severe lag in status monitoring results, which cannot meet the monitoring needs of devices starting and stopping, which require rapid response. Second, the network bandwidth and cloud resources are consumed in huge quantities. A large number of terminal devices continuously upload raw data, which puts enormous pressure on network bandwidth and cloud storage and computing resources. The system expansion cost is high and resources are wasted. Third, the computing load at the edge is too heavy, making it difficult to achieve real-time processing on low-cost, low-power microcontrollers. Although the second type of solution pushes some feature extraction tasks to the edge, it usually performs large-scale time-frequency analysis and high-dimensional feature construction on the edge. For example, it performs full-length or multi-overlapping window Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), Empirical Mode Decomposition (EMD), and Continuous Wavelet Transform (CWT) on multi-channel, long time series, and calculates time-domain, frequency-domain, and time-frequency domain combined features of more than ten or even dozens of dimensions on this basis, and even superimposes deep learning model inference. When these algorithm chains run on edge microcontroller units (MCUs), they place high demands on processor clock speed, storage space, and power consumption. As a result, the system is either forced to adopt high-performance, high-cost edge hardware, or it is difficult to achieve stable real-time processing on low-cost, low-power MCUs, thus limiting the feasibility of large-scale deployment of this type of technology in resource-constrained industrial environments. Fourth: The monitoring modes are singular and lack sensor self-diagnosis, resulting in insufficient robustness of diagnostic results. Most solutions only collect a single mechanical mode (such as vibration or current) or rely solely on acoustic signals, without building a multi-modal fusion mechanism for mechanical and acoustic signals at the system level, and also lacking methods for modeling and identifying the health status of the sensors themselves.When sensor coupling deteriorates, installation conditions change, or there is environmental noise interference, it is difficult to distinguish between "equipment malfunction" and "sensor malfunction", which can easily lead to false alarms or missed alarms. At the same time, single-mode monitoring is easily affected by operating condition fluctuations and noise, resulting in insufficient robustness and interpretability of monitoring results.

[0004] Therefore, how to design a monitoring method that can achieve stable real-time monitoring with low cost and low power consumption while ensuring monitoring accuracy has become an urgent problem to be solved. Summary of the Invention

[0005] Based on this, the present invention proposes an acoustic-vibration multimodal edge-cloud collaborative monitoring method based on adaptive dual entropy. This method extracts features from multimodal signal sample data to obtain multimodal features, achieving second-level delivery of window-level state features and triggering alarms. The cloud primarily performs multimodal fusion and self-diagnosis, significantly reducing overall end-to-end latency and meeting the real-time monitoring requirements for equipment start-up / shutdown identification and rapid response to abnormal operating conditions. Three-dimensional features are used to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices. The cloud aggregates and hierarchically retains the three-dimensional feature stream, only transmitting short acoustic segments as needed when a few abnormal events occur; the rest are recorded using structured features. This invention provides long-term data storage, significantly reducing the pressure on wireless link bandwidth and cloud storage and computing load. It allows for the connection of numerous devices and long-term online monitoring even in narrowband IoT environments, significantly improving scalability and cost control. It enables stable operation of the entire edge feature extraction process on a low-cost, low-power hardware platform, ensuring real-time performance and reliability during large-scale deployment in resource-constrained industrial environments. Based on multimodal features, it performs multimodal fusion discrimination to obtain multimodal fusion monitoring results, enabling stable identification of multiple states and improving robustness and discrimination accuracy in complex working conditions and high-noise environments. This invention enhances the stability and accuracy of intelligent monitoring of electromechanical equipment.

[0006] This invention proposes a method for collaborative monitoring of acoustic-vibration multimodal edge clouds based on adaptive dual entropy, comprising:

[0007] Multimodal signal sample data is collected and preprocessed, including acoustic signal sample data and vibration signal sample data;

[0008] Feature extraction is performed on multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features;

[0009] Multimodal fusion discrimination is performed based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators.

[0010] In summary, based on the aforementioned adaptive dual-entropy-based acoustic-vibration multimodal edge-cloud collaborative monitoring method, feature extraction is performed on multimodal signal sample data to obtain multimodal features. This enables the provision of window-level state features and alarm triggering within seconds. The cloud primarily performs multimodal fusion and self-diagnosis, significantly reducing overall end-to-end latency. This meets the real-time monitoring requirements for equipment start-up / stop identification and rapid response to abnormal operating conditions. Three-dimensional features are used to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices. The cloud aggregates and hierarchically retains the three-dimensional feature stream, only transmitting short acoustic segments as needed when a few abnormal events occur, while the rest are recorded and stored long-term using structured features. This invention significantly reduces the bandwidth pressure on wireless links and the cloud storage and computing load. It can still connect a large number of devices and achieve long-term online monitoring in narrowband IoT environments, significantly improving scalability and cost controllability. It enables the stable operation of the entire edge feature extraction process on a low-cost, low-power hardware platform, ensuring real-time performance and reliability when deployed on a large scale in resource-constrained industrial sites. It performs multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results, and can stably identify multi-level states. It improves robustness and discrimination accuracy in complex working conditions and high-noise environments. This invention improves the stability and accuracy of intelligent monitoring of electromechanical equipment. Specifically, the process involves collecting and preprocessing multimodal signal sample data, including acoustic and vibration signal sample data. Feature extraction is performed on the multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features. This enables the generation of window-level state features and alarm triggering within seconds. The cloud primarily performs multimodal fusion and self-diagnosis, significantly reducing overall end-to-end latency and meeting the real-time monitoring requirements for equipment start / stop identification and rapid response to abnormal operating conditions. Three-dimensional features are used to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices. The cloud aggregates and hierarchically retains the three-dimensional feature stream, only transmitting short acoustic segments as needed during a few abnormal events; the rest are stored in structured form. Long-term storage of feature records significantly reduces the bandwidth pressure on wireless links and the cloud storage and computing load. Even in narrowband IoT environments, it can still connect a large number of devices and achieve long-term online monitoring, significantly improving scalability and cost controllability. It enables stable operation of the entire edge feature extraction process on a low-cost, low-power hardware platform, ensuring real-time performance and reliability during large-scale deployment in resource-constrained industrial sites. Multimodal fusion and discrimination are performed based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion and discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators, which can stably identify multiple states and improve robustness and discrimination accuracy in complex working conditions and high-noise environments. This invention improves the stability and accuracy of intelligent monitoring of electromechanical equipment.

[0011] Furthermore, the step of acquiring multimodal signal sample data and performing preprocessing specifically includes:

[0012] The edge gateway receives acoustic signal sample data and vibration signal sample data from multiple acquisition nodes. For each device, the edge performs time alignment and segmentation on the signal sample data from the acoustic channel and vibration channel. The time alignment and segmentation are based on a unified window length and step size parameter, dividing the continuous time series into multiple analysis windows to generate corresponding acoustic data segments and vibration data segments, and assigning a timestamp and device identifier to each window.

[0013] Amplitude correction and bandwidth constraint are performed on the original acoustic signal sample data at the edge.

[0014] Furthermore, the step of extracting features from the multimodal signal sample data to obtain multimodal features specifically includes:

[0015] Adaptive time-domain discrete entropy feature extraction is performed on the preprocessed acoustic signal sample data to obtain acoustic time-domain entropy features;

[0016] The specific algorithm for adaptive time-domain discrete entropy feature extraction is as follows:

[0017] ,

[0018] ,

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] in, Indicates skewness, N Indicates the number of samples. i Indicates the sample ordinal number. Indicates the first i One acoustic signal sample data, This represents the mean. Standard deviation, log L The log-likelihood function represents the maximum likelihood value. c Indicates category, j Indicates the category ordinal number, Indicates the first j A count, AIC ( c () represents the optimal number of categories. Represents a discrete symbol sequence; `round` indicates that the mapped signal is discretized. Represents the logarithmic mapping of signal sample data. Represents the pattern probability distribution vector. m Indicates the embedding dimension. Indicates the delay time. ATDE represents the acoustic time-domain entropy characteristic, while ATDE represents the adaptive time-domain discrete entropy. X Indicates sample features, Indicates the number of patterns;

[0024] Frequency domain entropy features are extracted from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features.

[0025] Vibration intensity features are extracted from vibration signal sample data to obtain vibration intensity characteristics.

[0026] Furthermore, the step of extracting frequency domain entropy features from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features specifically includes:

[0027] The specific algorithm for frequency domain entropy feature extraction is as follows:

[0028] ,

[0029] ,

[0030] ,

[0031] ,

[0032] ,

[0033] in, Represents the time-frequency matrix. Indicates the time offset. This represents the frequency index, where M represents the frequency number and n represents the ordinal number. Represents acoustic signal sample data, j Indicates the category ordinal number, Represents the logarithmic energy spectrum matrix. Represents the matrix spectrum. This represents the grayscale matrix, and `round` indicates that the mapped signal has been discretized. Represents the acoustic frequency domain entropy characteristics. This represents the probability of occurrence of each gray level. The step of extracting frequency domain entropy features from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features specifically includes:

[0034] The specific algorithm for frequency domain entropy feature extraction is as follows:

[0035] ,

[0036] ,

[0037] ,

[0038] ,

[0039] ,

[0040] in, Represents the time-frequency matrix. Indicates the time offset. This represents the frequency index, where M represents the frequency number and n represents the ordinal number. Represents acoustic signal sample data, j Indicates the category ordinal number, Represents the logarithmic energy spectrum matrix. Represents the matrix spectrum. This represents the grayscale matrix, and `round` indicates that the mapped signal has been discretized. Represents the acoustic frequency domain entropy characteristics. This represents the probability of each gray level appearing.

[0041] Furthermore, the step of extracting vibration intensity features from the vibration signal sample data to obtain vibration intensity features specifically includes:

[0042] For the vibration data segment within the window, the root mean square value of the vibration acceleration is calculated at the edge to obtain the vibration intensity characteristics. The specific algorithm for these vibration intensity characteristics is as follows:

[0043] ,

[0044] in, Indicates the characteristics of vibration intensity. N This represents the number of samples, where n represents the ordinal number. This represents the vibration data segment in the vibration signal sample data.

[0045] Furthermore, the step of performing multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results specifically includes:

[0046] The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are sorted and grouped according to the device identifier and timestamp to construct a multimodal time series for each device.

[0047] Based on vibration intensity characteristics, the start-up, shutdown, and operating conditions of the equipment are identified and mechanical condition labels are applied.

[0048] If the vibration intensity characteristics are all below the lower limit threshold of vibration intensity during the monitoring period, the target equipment will be marked as shut down.

[0049] If the vibration intensity characteristics are higher than the vibration intensity operating threshold throughout the monitoring period, the target equipment will be marked as being in a stable operating state.

[0050] When the vibration intensity characteristic crosses the lower limit threshold of vibration intensity from low to high or crosses the operating threshold of vibration intensity from high to low within the monitoring period, and the rate of change exceeds the preset threshold, the target equipment will be marked as either in start-up state or in shutdown transition state, respectively.

[0051] Within the time window set of stable operating conditions, the mean and standard deviation of acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are calculated. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are then normalized to obtain normalized feature vectors. The mean of the normalized feature vectors is then calculated to obtain the center vector of the normal operating mode.

[0052] Based on the normalized feature vector and the normal operation mode center vector, a weighted multimodal distance index is calculated. The specific algorithm for calculating the weighted multimodal distance index is as follows:

[0053] ,

[0054] in, This represents the weighted multimodal distance index. , , These represent the acoustic time-domain entropy feature weights, acoustic frequency-domain entropy feature weights, and vibration intensity feature weights, respectively. , , Let represent the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively. , , These represent the normal operating mode center vectors of the normalized acoustic time-domain entropy eigenvectors, the normalized acoustic frequency-domain entropy eigenvectors, and the normalized vibration intensity eigenvectors, respectively.

[0055] The mean and standard deviation of the weighted multimodal distance index are calculated to obtain the thresholds for mild and severe anomalies. The specific algorithms for calculating the mild and severe anomaly thresholds are as follows:

[0056] ,

[0057] in, Indicates the threshold for mild abnormalities. Indicates the threshold for severe anomalies. This represents the mean of the weighted multimodal distance index. The standard deviation of the weighted multimodal distance index is represented by... and These represent the standard deviation weights of the mild and severe anomaly thresholds, respectively.

[0058] The system is classified and judged based on mechanical condition labels and weighted multimodal distance indicators to obtain multimodal fusion monitoring results;

[0059] When the mechanical condition label is "shutdown" and the acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics are all within the normal shutdown range, the shutdown is considered normal.

[0060] When the mechanical condition label is in a stable operating state and the weighted multimodal distance index is less than the mild anomaly threshold, it is judged to be operating normally;

[0061] When the mechanical operating condition label is in a stable operating state, and the weighted multimodal distance index is greater than or equal to the mild anomaly threshold and less than the severe anomaly threshold, the operation is judged to be suspicious.

[0062] When the mechanical operating condition label is in a stable operating state and the weighted multimodal distance index is greater than or equal to the severe anomaly threshold, it is judged as an operating anomaly.

[0063] Furthermore, after the step of classifying and judging based on mechanical condition labels and weighted multimodal distance indicators to obtain multimodal fusion monitoring results, the method further includes:

[0064] Perform equipment self-diagnosis;

[0065] When the target equipment is shut down, consistency diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. If the vibration signal has stopped but the acoustic signal is still significantly active within the monitoring period, it is determined that the current acoustic measurement point is affected by environmental noise interference or has a poor coupling problem.

[0066] Under the target device's start-up and stop states, consistency diagnosis is performed based on acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics. For each start-up and stop event, a representative window is selected before and after the conversion, and the changes in acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics are calculated to construct a comprehensive response amplitude. Based on the comprehensive response amplitude, if the vibration signal feedback device is in a start-up and stop state during the monitoring period, but the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics do not change, it is determined that the acoustic sensor has a serious decoupling, damage, or signal link failure problem. The specific algorithm for the comprehensive response amplitude is as follows:

[0067] ,

[0068] ,

[0069] in, This represents the change in acoustic temporal entropy characteristics. This represents the change in acoustic frequency domain entropy characteristics. and Let represent the acoustic temporal entropy characteristics of the window after start / stop and the acoustic temporal entropy characteristics of the window before start / stop, respectively. and These represent the acoustic frequency domain entropy characteristics of the window after start / stop and the acoustic frequency domain entropy characteristics of the window before start / stop, respectively. Indicates the overall response magnitude;

[0070] Under stable operating conditions of the target equipment, correlation degradation diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. An acoustic comprehensive complexity index is constructed, and the baseline correlation coefficient for the historical healthy phase and the correlation coefficient for the current stable operating phase are calculated and compared. If the changes in vibration intensity characteristics during the monitoring period cannot be reflected in the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics, it is determined that the acoustic sensor performance is slowly degrading or the coupling conditions are deteriorating over a long period. The specific algorithm for the acoustic comprehensive complexity index is as follows:

[0071] ,

[0072] Among them, U( j ) represents the acoustic complexity index. , These represent the complexity weights of the acoustic time-domain entropy features and the acoustic frequency-domain entropy features, respectively. Represents the acoustic temporal entropy characteristics. This represents the acoustic frequency domain entropy characteristics.

[0073] This invention proposes an acoustic-vibration multimodal edge cloud collaborative monitoring system based on adaptive dual entropy, comprising:

[0074] The acquisition module is used to acquire multimodal signal sample data and perform preprocessing. The multimodal signal sample data includes acoustic signal sample data and vibration signal sample data.

[0075] The feature extraction module is used to extract features from multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features.

[0076] The discrimination module is used to perform multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators.

[0077] The present invention also provides a storage medium that stores one or more programs, which, when executed by a processor, implement the above-described adaptive dual-entropy-based acoustic-vibration multimodal edge cloud collaborative monitoring method.

[0078] The present invention also provides a computer device, the computer device including a memory and a processor, wherein:

[0079] The memory is used to store computer programs;

[0080] When the processor executes the computer program stored in the memory, it implements the acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy as described above. Attached Figure Description

[0081] Figure 1 This is a flowchart of the acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy proposed in the first embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram of the structure of the acoustic-vibration multimodal edge cloud collaborative monitoring system based on adaptive dual entropy proposed in the second embodiment of the present invention.

[0083] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0084] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0085] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0086] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0087] Please see Figure 1 The diagram shows a flowchart of the acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy proposed in the first embodiment of the present invention. This acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy includes steps S01 to S03, wherein:

[0088] Step S01: Acquire multimodal signal sample data and perform preprocessing;

[0089] It should be noted that in this embodiment, the multimodal signal sample data includes acoustic signal sample data and vibration signal sample data. The edge gateway receives acoustic signal sample data and vibration signal sample data from multiple acquisition nodes. For each device, the edge performs time alignment and segmentation on the signal sample data from the acoustic channel and vibration channel. The time alignment and segmentation are based on a unified window length and step size parameter, dividing the continuous time series into multiple analysis windows to generate corresponding acoustic data segments and vibration data segments, and assigning a timestamp and device identifier to each window.

[0090] Amplitude correction and bandwidth constraint are performed on the original acoustic signal sample data at the edge.

[0091] In this embodiment, an intelligent sensing and acquisition module is deployed at the field acquisition node to synchronously acquire the acoustic / acoustic emission signals and vibration signals of the device under unattended conditions, and to complete front-end conditioning, basic feature calculation, event triggering, data caching, and low-power management. This module includes at least an acoustic acquisition sub-channel and a vibration acquisition sub-channel.

[0092] The acoustic acquisition subchannel uses a RAEM2 acoustic emission sensor to collect operating sound waves or acoustic emission signals from the surface of the equipment housing, base, or pipe. After front-end amplification and bandpass filtering, it outputs a net signal that meets the requirements of analog-to-digital conversion. The vibration acquisition subchannel installs a MEMS accelerometer on the equipment housing, base, or bearing housing to collect vibration acceleration time-domain signals. After front-end amplification, anti-aliasing filtering, and amplitude limiting, it outputs a sampleable net vibration signal. Based on the above hardware, the intelligent sensing and acquisition module operates according to the following steps: 1. Signal acquisition and front-end conditioning: The acoustic acquisition subchannel acquires sound wave / acoustic emission signals from the equipment surface, and the vibration acquisition subchannel simultaneously acquires vibration acceleration signals. The two signals are respectively subjected to analog conditioning such as front-end amplification, bandpass filtering, and amplitude limiting to remove DC drift and interference components outside the operating frequency band, outputting a net signal that meets the input range of the analog-to-digital converter; 2. Configurable sampling and segmented acquisition: The acquisition unit processes the acoustic net signal and vibration net signal according to parameters such as sampling rate, window length, and sampling period sent by the host computer or cloud. Synchronous sampling divides the continuous time series into several data segments according to a set window length, forming acoustic data segments and corresponding vibration data segments. A timestamp and device identifier are added to each data segment to provide a unified data format for subsequent feature extraction by the edge gateway. 3. End-side basic feature calculation: The data acquisition unit calculates acoustic basic indicators with low computational complexity, such as amplitude, root mean square (RMS), power, and average sound pressure level (ASL), for each acoustic data segment in real time. Simultaneously, it calculates basic features such as the root mean square of vibration for the corresponding vibration data segment to quickly reflect the current vibration intensity level and rough operating status. 4. Event triggering and window marking: The collector performs threshold judgment on acoustic and vibration basic features based on preset thresholds or change rate rules. When the acoustic or vibration features of a data segment exceed the normal range or undergo drastic changes in a short period of time, the data segment is marked as an event window. A short-term raw acoustic waveform segment can be selectively captured before and after the event for further detailed analysis at the edge gateway or in the cloud. 5. Result caching and breakpoint resume support: The collector performs "basic features + event marking" on data from each time period. The data acquisition unit stores the information in a structured manner, including "data + timestamp + device ID". For data segments marked as events, the corresponding short acoustic waveform segments are attached. In the event of wireless link interruption or reporting failure, the above data is temporarily stored in the local buffer and a retransmission queue is recorded. After communication is restored, the data is retransmitted in chronological order to ensure data integrity. 6. Low power consumption operation and self-test: The acquisition unit adopts a "sleep-wake-acquisition-reporting-re-sleep" working mode. It wakes up periodically according to the set duty cycle to complete signal acquisition and data reporting. At other times, it enters a low power sleep state to reduce energy consumption during long-term monitoring.Simultaneously, the data acquisition unit performs self-checks on operating parameters such as battery level, device temperature, and open / short circuits in the acoustic / vibration acquisition channels. If insufficient power or channel abnormalities are detected, a local self-check alarm is generated and reported along with the basic characteristics. 7. Wireless parameter uplink and remote configuration: The data acquisition unit primarily uses lightweight parameter messages to report data including "acoustic basic characteristics, vibration basic characteristics, event markers, time information, and device identifiers" to the edge gateway or cloud platform via a LoRa wireless link. When more detailed analysis of events is required, short-time acoustic waveform clips can be uploaded. The system supports remote distribution of configuration parameters such as sampling rate, window length, threshold, and reporting strategy. The data acquisition unit receives and updates its local configuration, which then takes effect automatically, thereby enabling remote adjustment and maintenance of on-site monitoring strategies.

[0093] Step S02: Extract features from the multimodal signal sample data to obtain multimodal features;

[0094] It should be noted that in this embodiment, the multimodal features include acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features. Adaptive time-domain discrete entropy feature extraction is performed on the preprocessed acoustic signal sample data to obtain acoustic time-domain entropy features.

[0095] The specific algorithm for adaptive time-domain discrete entropy feature extraction is as follows:

[0096] ,

[0097] ,

[0098] ,

[0099] ,

[0100] ,

[0101] ,

[0102] in, Indicates skewness, N Indicates the number of samples. i Indicates the sample ordinal number. Indicates the first i One acoustic signal sample data, This represents the mean. Standard deviation, log L The log-likelihood function represents the maximum likelihood value. c Indicates category, j Indicates the category ordinal number, Indicates the first j A count, AIC ( c() represents the optimal number of categories. Represents a discrete symbol sequence; `round` indicates that the mapped signal is discretized. Represents the logarithmic mapping of signal sample data. Represents the pattern probability distribution vector. m Indicates the embedding dimension. Indicates the delay time. ATDE represents the acoustic time-domain entropy characteristic, while ATDE represents the adaptive time-domain discrete entropy. X Indicates sample features, Indicates the number of patterns;

[0103] Frequency domain entropy features are extracted from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features.

[0104] Vibration intensity features are extracted from vibration signal sample data to obtain vibration intensity characteristics.

[0105] The specific algorithm for frequency domain entropy feature extraction is as follows:

[0106] ,

[0107] ,

[0108] ,

[0109] ,

[0110] ,

[0111] in, Represents the time-frequency matrix. Indicates the time offset. This represents the frequency index, where M represents the frequency number and n represents the ordinal number. Represents acoustic signal sample data, j Indicates the category ordinal number, Represents the logarithmic energy spectrum matrix. Represents the matrix spectrum. This represents the grayscale matrix, and `round` indicates that the mapped signal has been discretized. Represents the acoustic frequency domain entropy characteristics. This represents the probability of each gray level appearing.

[0112] For the vibration data segment within the window, the root mean square value of the vibration acceleration is calculated at the edge to obtain the vibration intensity characteristics. The specific algorithm for these vibration intensity characteristics is as follows:

[0113] ,

[0114] in, Indicates the characteristics of vibration intensity. N This represents the number of samples, where n represents the ordinal number. This represents the vibration data segment in the vibration signal sample data.

[0115] Step S03: Perform multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results;

[0116] It should be noted that in this embodiment, the multimodal fusion discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators. The acoustic time domain entropy features, acoustic frequency domain entropy features, and vibration intensity features are sorted and grouped according to the equipment identifier and timestamp to construct the multimodal time series of each device.

[0117] Based on vibration intensity characteristics, the start-up, shutdown, and operating conditions of the equipment are identified and mechanical condition labels are applied.

[0118] If the vibration intensity characteristics are all below the lower limit threshold of vibration intensity during the monitoring period, the target equipment will be marked as shut down.

[0119] If the vibration intensity characteristics are higher than the vibration intensity operating threshold throughout the monitoring period, the target equipment will be marked as being in a stable operating state.

[0120] When the vibration intensity characteristic crosses the lower limit threshold of vibration intensity from low to high or crosses the operating threshold of vibration intensity from high to low within the monitoring period, and the rate of change exceeds the preset threshold, the target equipment will be marked as either in start-up state or in shutdown transition state, respectively.

[0121] Within the time window set of stable operating conditions, the mean and standard deviation of acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are calculated. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are then normalized to obtain normalized feature vectors. The mean of the normalized feature vectors is then calculated to obtain the center vector of the normal operating mode.

[0122] Based on the normalized feature vector and the normal operation mode center vector, a weighted multimodal distance index is calculated. The specific algorithm for calculating the weighted multimodal distance index is as follows:

[0123] ,

[0124] in, This represents the weighted multimodal distance index. , , These represent the acoustic time-domain entropy feature weights, acoustic frequency-domain entropy feature weights, and vibration intensity feature weights, respectively. , , Let represent the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively. , , These represent the normal operating mode center vectors of the normalized acoustic time-domain entropy eigenvectors, the normalized acoustic frequency-domain entropy eigenvectors, and the normalized vibration intensity eigenvectors, respectively.

[0125] The mean and standard deviation of the weighted multimodal distance index are calculated to obtain the thresholds for mild and severe anomalies. The specific algorithms for calculating the mild and severe anomaly thresholds are as follows:

[0126] ,

[0127] in, Indicates the threshold for mild abnormalities. Indicates the threshold for severe anomalies. This represents the mean of the weighted multimodal distance index. The standard deviation of the weighted multimodal distance index is represented by... and These represent the standard deviation weights of the mild and severe anomaly thresholds, respectively.

[0128] The system is classified and judged based on mechanical condition labels and weighted multimodal distance indicators to obtain multimodal fusion monitoring results;

[0129] When the mechanical condition label is "shutdown" and the acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics are all within the normal shutdown range, the shutdown is considered normal.

[0130] When the mechanical condition label is in a stable operating state and the weighted multimodal distance index is less than the mild anomaly threshold, it is judged to be operating normally;

[0131] When the mechanical operating condition label is in a stable operating state, and the weighted multimodal distance index is greater than or equal to the mild anomaly threshold and less than the severe anomaly threshold, the operation is judged to be suspicious.

[0132] When the mechanical operating condition label is in a stable operating state and the weighted multimodal distance index is greater than or equal to the severe anomaly threshold, it is judged as an operating anomaly.

[0133] Perform equipment self-diagnosis;

[0134] When the target equipment is shut down, consistency diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. If the vibration signal has stopped but the acoustic signal is still significantly active within the monitoring period, it is determined that the current acoustic measurement point is affected by environmental noise interference or has a poor coupling problem.

[0135] Under the target device's start-up and stop states, consistency diagnosis is performed based on acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics. For each start-up and stop event, a representative window is selected before and after the conversion, and the changes in acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics are calculated to construct a comprehensive response amplitude. Based on the comprehensive response amplitude, if the vibration signal feedback device is in a start-up and stop state during the monitoring period, but the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics do not change, it is determined that the acoustic sensor has a serious decoupling, damage, or signal link failure problem. The specific algorithm for the comprehensive response amplitude is as follows:

[0136] ,

[0137] ,

[0138] in, This represents the change in acoustic temporal entropy characteristics. This represents the change in acoustic frequency domain entropy characteristics. and Let represent the acoustic temporal entropy characteristics of the window after start / stop and the acoustic temporal entropy characteristics of the window before start / stop, respectively. and These represent the acoustic frequency domain entropy characteristics of the window after start / stop and the acoustic frequency domain entropy characteristics of the window before start / stop, respectively. Indicates the overall response magnitude;

[0139] Under stable operating conditions of the target equipment, correlation degradation diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. An acoustic comprehensive complexity index is constructed, and the baseline correlation coefficient for the historical healthy phase and the correlation coefficient for the current stable operating phase are calculated and compared. If the changes in vibration intensity characteristics during the monitoring period cannot be reflected in the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics, it is determined that the acoustic sensor performance is slowly degrading or the coupling conditions are deteriorating over a long period. The specific algorithm for the acoustic comprehensive complexity index is as follows:

[0140] ,

[0141] Among them, U( j ) represents the acoustic complexity index. , These represent the complexity weights of the acoustic time-domain entropy features and the acoustic frequency-domain entropy features, respectively. Represents the acoustic temporal entropy characteristics. This represents the acoustic frequency domain entropy characteristics.

[0142] Based on multimodal fusion discrimination and sensor self-diagnosis, this invention centrally deploys visualization and alarm display on a mobile application. The cloud platform is only responsible for storing multimodal features, status tags, and alarm records, and provides query and subscription capabilities to mobile terminals through a unified data service interface. The mobile visualization and alarm display submodule includes at least the following functions: 1. Time series trend chart: The mobile terminal obtains the multimodal features and status tags of each device within a selected time range from the cloud, plots historical curves, and overlays mechanical condition tags and device status tags to intuitively present the start-up and shutdown process and abnormal intervals. 2. Sensor health status view: Based on the sensor health status tag Ssen(j) output from the cloud, the mobile terminal displays the current health status of each acoustic measurement point using color, icon, or level indicators, and provides a timeline playback function, enabling maintenance personnel to view the evolution of sensor health status at different time periods to distinguish between equipment problems and sensor problems. 3. Alarm list and details: The mobile terminal receives or pulls the time period marked as "RUN_ABNORMAL" by the multimodal fusion discrimination module and the corresponding abnormal records from the cloud to form an alarm list. For any alarm entry, users can access the details interface to view a summary of multimodal characteristics for the associated time period (including typical values ​​and trends), multimodal distance indicators, and specific alarm reasons and recommendations determined by the cloud. This facilitates maintenance personnel's understanding of the alarm basis and enables them to make quick decisions. Through the aforementioned mobile visualization and alarm display submodules, this invention achieves mobile presentation of industrial equipment operating status and sensor health status, as well as closed-loop alarm management, without increasing the complexity of the cloud interface. It supports maintenance personnel in obtaining key monitoring information in real time on-site or remote terminals.

[0143] In summary, based on the aforementioned adaptive dual-entropy-based acoustic-vibration multimodal edge-cloud collaborative monitoring method, feature extraction is performed on multimodal signal sample data to obtain multimodal features. This enables the provision of window-level state features and alarm triggering within seconds. The cloud primarily performs multimodal fusion and self-diagnosis, significantly reducing overall end-to-end latency. This meets the real-time monitoring requirements for equipment start-up / stop identification and rapid response to abnormal operating conditions. Three-dimensional features are used to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices. The cloud aggregates and hierarchically retains the three-dimensional feature stream, only transmitting short acoustic segments as needed when a few abnormal events occur, while the rest are recorded and stored long-term using structured features. This invention significantly reduces the bandwidth pressure on wireless links and the cloud storage and computing load. It can still connect a large number of devices and achieve long-term online monitoring in narrowband IoT environments, significantly improving scalability and cost controllability. It enables the stable operation of the entire edge feature extraction process on a low-cost, low-power hardware platform, ensuring real-time performance and reliability when deployed on a large scale in resource-constrained industrial sites. It performs multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results, and can stably identify multi-level states. It improves robustness and discrimination accuracy in complex working conditions and high-noise environments. This invention improves the stability and accuracy of intelligent monitoring of electromechanical equipment. Specifically, the process involves collecting and preprocessing multimodal signal sample data, including acoustic and vibration signal sample data. Feature extraction is performed on the multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features. This enables the generation of window-level state features and alarm triggering within seconds. The cloud primarily performs multimodal fusion and self-diagnosis, significantly reducing overall end-to-end latency and meeting the real-time monitoring requirements for equipment start / stop identification and rapid response to abnormal operating conditions. Three-dimensional features are used to replace the uplink of complete acoustic waveforms and high-dimensional time-frequency matrices. The cloud aggregates and hierarchically retains the three-dimensional feature stream, only transmitting short acoustic segments as needed during a few abnormal events; the rest are stored in structured form. Long-term storage of feature records significantly reduces the bandwidth pressure on wireless links and the cloud storage and computing load. Even in narrowband IoT environments, it can still connect a large number of devices and achieve long-term online monitoring, significantly improving scalability and cost controllability. It enables stable operation of the entire edge feature extraction process on a low-cost, low-power hardware platform, ensuring real-time performance and reliability during large-scale deployment in resource-constrained industrial sites. Multimodal fusion and discrimination are performed based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion and discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators, which can stably identify multiple states and improve robustness and discrimination accuracy in complex working conditions and high-noise environments. This invention improves the stability and accuracy of intelligent monitoring of electromechanical equipment.

[0144] Please see Figure 2The figure shown is a schematic diagram of the acoustic-vibration multimodal edge cloud collaborative monitoring system based on adaptive dual entropy proposed in the second embodiment of the present invention. The system includes:

[0145] The acquisition module 10 is used to acquire multimodal signal sample data and perform preprocessing. The multimodal signal sample data includes acoustic signal sample data and vibration signal sample data.

[0146] The feature extraction module 20 is used to extract features from multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features.

[0147] The discrimination module 30 is used to perform multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators.

[0148] The present invention also proposes a computer storage medium storing one or more programs that, when executed by a processor, implement the aforementioned acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy.

[0149] The present invention also proposes a computer device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the above-mentioned acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy.

[0150] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0151] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0152] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0153] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0154] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An adaptive dual-entropy-based acoustic-vibration multi-modal edge cloud collaborative monitoring method, characterized in that, include: Multimodal signal sample data is collected and preprocessed, including acoustic signal sample data and vibration signal sample data; Feature extraction is performed on multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features; The step of extracting features from multimodal signal sample data to obtain multimodal features specifically includes: Adaptive time-domain discrete entropy feature extraction is performed on the preprocessed acoustic signal sample data to obtain acoustic time-domain entropy features; The specific algorithm for adaptive time-domain discrete entropy feature extraction is as follows: ， ， ， ， ， ， in, Indicates skewness, N Indicates the number of samples. i Indicates the sample ordinal number. Indicates the first i One acoustic signal sample data, This represents the mean. Standard deviation, log L The log-likelihood function represents the maximum likelihood value. c Indicates category, j Indicates the category ordinal number, Indicates the first j A count, AIC ( c () represents the optimal number of categories. Represents a discrete symbol sequence; `round` indicates that the mapped signal is discretized. Represents the logarithmic mapping of signal sample data. Represents the pattern probability distribution vector. m Indicates the embedding dimension. Indicates the delay time. ATDE represents the acoustic time-domain entropy characteristic, while ATDE represents the adaptive time-domain discrete entropy. X Indicates sample features, Indicates the number of patterns; Frequency domain entropy features are extracted from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features. Vibration intensity features are extracted from vibration signal sample data to obtain vibration intensity characteristics; Multimodal fusion and discrimination are performed based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion and discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are sorted and grouped according to the device identifier and timestamp to construct a multimodal time series for each device. Based on vibration intensity characteristics, the start-up, shutdown, and operating conditions of the equipment are identified and mechanical condition labels are applied. Within the time window set of stable operating conditions, the mean and standard deviation of acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are calculated. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are then normalized to obtain normalized feature vectors. The mean of the normalized feature vectors is then calculated to obtain the center vector of the normal operating mode. Based on the normalized feature vector and the normal operation mode center vector, a weighted multimodal distance index is calculated. The specific algorithm for calculating the weighted multimodal distance index is as follows: ， in, This represents the weighted multimodal distance index. , , These represent the acoustic time-domain entropy feature weights, acoustic frequency-domain entropy feature weights, and vibration intensity feature weights, respectively. , , Let represent the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively. , , These represent the normal operating mode center vectors of the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively.

2. The acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy according to claim 1, characterized in that, The steps of acquiring and preprocessing multimodal signal sample data specifically include: The edge gateway receives acoustic signal sample data and vibration signal sample data from multiple acquisition nodes. For each device, the edge performs time alignment and segmentation on the signal sample data from the acoustic channel and vibration channel. The time alignment and segmentation are based on a unified window length and step size parameter, dividing the continuous time series into multiple analysis windows to generate corresponding acoustic data segments and vibration data segments, and assigning a timestamp and device identifier to each window. Amplitude correction and bandwidth constraint are performed on the original acoustic signal sample data at the edge.

3. The acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy according to claim 1, characterized in that, The step of extracting frequency domain entropy features from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features specifically includes: The specific algorithm for frequency domain entropy feature extraction is as follows: ， ， ， ， ， in, Represents the time-frequency matrix. Indicates the time offset. This represents the frequency index, where M represents the frequency number and n represents the ordinal number. Represents acoustic signal sample data, j Indicates the category ordinal number, Represents the logarithmic energy spectrum matrix. Represents the matrix spectrum. This represents the grayscale matrix, and `round` indicates that the mapped signal has been discretized. Represents the acoustic frequency domain entropy characteristics. This represents the probability of each gray level appearing.

4. The acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy according to claim 1, characterized in that, The step of extracting vibration intensity features from vibration signal sample data to obtain vibration intensity features specifically includes: For the vibration data segment within the window, the root mean square value of the vibration acceleration is calculated at the edge to obtain the vibration intensity characteristics. The specific algorithm for these vibration intensity characteristics is as follows: ， in, Indicates the characteristics of vibration intensity. N This represents the number of samples, where n represents the ordinal number. This represents the vibration data segment in the vibration signal sample data.

5. The acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy according to claim 1, characterized in that, The step of performing multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results specifically includes: If the vibration intensity characteristics are all below the lower limit threshold of vibration intensity during the monitoring period, the target equipment will be marked as shut down. If the vibration intensity characteristics are higher than the vibration intensity operating threshold throughout the monitoring period, the target equipment will be marked as being in a stable operating state. When the vibration intensity characteristic crosses the lower limit threshold of vibration intensity from low to high or crosses the operating threshold of vibration intensity from high to low within the monitoring period, and the rate of change exceeds the preset threshold, the target equipment will be marked as either in start-up state or in shutdown transition state, respectively. The mean and standard deviation of the weighted multimodal distance index are calculated to obtain the thresholds for mild and severe anomalies. The specific algorithms for calculating the mild and severe anomaly thresholds are as follows: ， ， ， in, Indicates the threshold for mild abnormalities. Indicates the threshold for severe anomalies. This represents the mean of the weighted multimodal distance index. The standard deviation of the weighted multimodal distance index is represented by... and These represent the standard deviation weights of the mild and severe anomaly thresholds, respectively. The system is classified and judged based on mechanical condition labels and weighted multimodal distance indicators to obtain multimodal fusion monitoring results; When the mechanical condition label is "shutdown" and the acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics are all within the normal shutdown range, the shutdown is considered normal. When the mechanical condition label is in a stable operating state and the weighted multimodal distance index is less than the mild anomaly threshold, it is judged to be operating normally; When the mechanical operating condition label is in a stable operating state, and the weighted multimodal distance index is greater than or equal to the mild anomaly threshold and less than the severe anomaly threshold, the operation is judged to be suspicious. When the mechanical operating condition label is in a stable operating state and the weighted multimodal distance index is greater than or equal to the severe anomaly threshold, it is judged as an operating anomaly.

6. The acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy according to claim 5, characterized in that, The step of classifying and judging based on mechanical condition labels and weighted multimodal distance indicators to obtain multimodal fusion monitoring results further includes: Perform equipment self-diagnosis; When the target equipment is shut down, consistency diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. If the vibration signal has stopped but the acoustic signal is still significantly active within the monitoring period, it is determined that the current acoustic measurement point is affected by environmental noise interference or has a poor coupling problem. Under the target device's start-up and stop states, consistency diagnosis is performed based on acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics. For each start-up and stop event, a representative window is selected before and after the conversion, and the changes in acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics are calculated to construct a comprehensive response amplitude. Based on the comprehensive response amplitude, if the vibration signal feedback device is in a start-up and stop state during the monitoring period, but the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics do not change, it is determined that the acoustic sensor has a serious decoupling, damage, or signal link failure problem. The specific algorithm for the comprehensive response amplitude is as follows: ， ， ， in, This represents the change in acoustic temporal entropy characteristics. This represents the change in acoustic frequency domain entropy characteristics. and Let represent the acoustic temporal entropy characteristics of the window after start / stop and the acoustic temporal entropy characteristics of the window before start / stop, respectively. and These represent the acoustic frequency domain entropy characteristics of the window after start / stop and the acoustic frequency domain entropy characteristics of the window before start / stop, respectively. Indicates the overall response magnitude; Under stable operating conditions of the target equipment, correlation degradation diagnosis is performed based on acoustic time-domain entropy characteristics, acoustic frequency-domain entropy characteristics, and vibration intensity characteristics. An acoustic comprehensive complexity index is constructed, and the baseline correlation coefficient for the historical healthy phase and the correlation coefficient for the current stable operating phase are calculated and compared. If the changes in vibration intensity characteristics during the monitoring period cannot be reflected in the acoustic time-domain entropy characteristics and acoustic frequency-domain entropy characteristics, it is determined that the acoustic sensor performance is slowly degrading or the coupling conditions are deteriorating over a long period. The specific algorithm for the acoustic comprehensive complexity index is as follows: ， Among them, U( j ) represents the acoustic complexity index. , These represent the complexity weights of the acoustic time-domain entropy features and the acoustic frequency-domain entropy features, respectively. Represents the acoustic temporal entropy characteristics. This represents the acoustic frequency domain entropy characteristics.

7. A collaborative monitoring system for acoustic-vibration multimodal edge clouds based on adaptive dual entropy, characterized in that, include: The acquisition module is used to acquire multimodal signal sample data and perform preprocessing. The multimodal signal sample data includes acoustic signal sample data and vibration signal sample data. The feature extraction module is used to extract features from multimodal signal sample data to obtain multimodal features, including acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features. The step of extracting features from multimodal signal sample data to obtain multimodal features specifically includes: Adaptive time-domain discrete entropy feature extraction is performed on the preprocessed acoustic signal sample data to obtain acoustic time-domain entropy features; The specific algorithm for adaptive time-domain discrete entropy feature extraction is as follows: ， ， ， ， ， ， in, Indicates skewness, N Indicates the number of samples. i Indicates the sample ordinal number. Indicates the first i One acoustic signal sample data, This represents the mean. Standard deviation, log L The log-likelihood function represents the maximum likelihood value. c Indicates category, j Indicates the category ordinal number, Indicates the first j A count, AIC ( c () represents the optimal number of categories. Represents a discrete symbol sequence; `round` indicates that the mapped signal is discretized. Represents the logarithmic mapping of signal sample data. Represents the pattern probability distribution vector. m Indicates the embedding dimension. Indicates the delay time. ATDE represents the acoustic time-domain entropy characteristic, while ATDE represents the adaptive time-domain discrete entropy. X Indicates sample features, Indicates the number of patterns; Frequency domain entropy features are extracted from the preprocessed acoustic signal sample data to obtain acoustic frequency domain entropy features. Vibration intensity features are extracted from vibration signal sample data to obtain vibration intensity characteristics; The discrimination module is used to perform multimodal fusion discrimination based on multimodal features to obtain multimodal fusion monitoring results. The multimodal fusion discrimination is a hierarchical discrimination based on mechanical condition labels and weighted multimodal distance indicators. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are sorted and grouped according to the device identifier and timestamp to construct a multimodal time series for each device. Based on vibration intensity characteristics, the start-up, shutdown, and operating conditions of the equipment are identified and mechanical condition labels are applied. Within the time window set of stable operating conditions, the mean and standard deviation of acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are calculated. The acoustic time-domain entropy features, acoustic frequency-domain entropy features, and vibration intensity features are then normalized to obtain normalized feature vectors. The mean of the normalized feature vectors is then calculated to obtain the center vector of the normal operating mode. Based on the normalized feature vector and the normal operation mode center vector, a weighted multimodal distance index is calculated. The specific algorithm for calculating the weighted multimodal distance index is as follows: ， in, This represents the weighted multimodal distance index. , , These represent the acoustic time-domain entropy feature weights, acoustic frequency-domain entropy feature weights, and vibration intensity feature weights, respectively. , , Let represent the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively. , , These represent the normal operating mode center vectors of the normalized acoustic time-domain entropy eigenvector, the normalized acoustic frequency-domain entropy eigenvector, and the normalized vibration intensity eigenvector, respectively.

8. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the acoustic-vibration multimodal edge cloud collaborative monitoring method based on any one of claims 1-6.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the acoustic-vibration multimodal edge cloud collaborative monitoring method based on adaptive dual entropy as described in any one of claims 1-6.

Images