Device failure warning method, system and storage medium under multi-source Internet of Things perception

Abstract

The present invention provides a device fault warning method, system, and storage medium under multi-source Internet of Things perception, relating to the field of device fault warning technology. The method establishes a fault propagation rate model by acquiring the spatial position coordinates of multiple sensing nodes in a multi-source Internet of Things sensing device; calls this model to calculate the propagation delay index of the sensing node pair, outputs a propagation delay index matrix, outputs a fault response time anchor point in combination with the fault response time of the sensing node, performs time series delay calibration on the multiple sensing nodes, outputs a calibrated time series, extracts calibrated multi-source sensing data for fault type identification, and outputs a fault warning signal. The present invention solves the technical problem of insufficient data fusion accuracy and insufficient fault warning accuracy and timeliness due to differences in the response time of multi-source sensing nodes in the prior art, thereby achieving the technical effect of realizing multi-source sensing data time series delay calibration, thereby improving the accuracy and timeliness of fault warning.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment failure warning technology, and in particular to an equipment failure warning method, system and storage medium under multi-source Internet of Things perception. Background Art

[0002] With the continuous development of Internet of Things (IoT) technology, device fault warning based on multi-source IoT sensing has gradually become an important technical means to ensure the safe and reliable operation of industrial equipment. In an IoT sensing system, multiple sensing nodes are deployed at key locations of equipment to collect real-time operational data. Existing IoT sensing-based device fault warning methods generally focus on anomaly detection in the data of a single sensing node, or simple preliminary integration and feature extraction of multi-source sensing data to identify potential fault risks. However, existing methods ignore the differences in response time among multi-source sensing nodes and fail to fully consider the propagation patterns of each sensing node under spatial distribution and physical topology. This makes it difficult to accurately match the temporal relationships of the data during data fusion, resulting in insufficient data fusion accuracy. This multi-node temporal deviation easily causes fault feature aliasing, reducing the accuracy of multimodal data fusion, thereby affecting the timeliness and reliability of fault warnings and limiting the effective use of multi-source sensing data for equipment health status perception and fault warning. Summary of the Invention

[0003] The present invention provides a device fault warning method, system and storage medium under multi-source Internet of Things perception, which solves the technical problems in the prior art caused by differences in the response time of multi-source perception nodes, resulting in insufficient data fusion accuracy and insufficient fault warning accuracy and timeliness, and achieves the technical effect of realizing multi-source perception data timing delay calibration, thereby improving the accuracy and timeliness of fault warning.

[0004] In view of the above problems, on the one hand, the present invention provides a device fault warning method under multi-source Internet of Things perception, the method comprising: obtaining multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source Internet of Things perception device; establishing a fault propagation rate model based on the multiple spatial position coordinates; combining the multiple sensing nodes to obtain sensing node pairs, calling the fault propagation rate model to calculate the propagation delay index of each sensing node pair, and outputting a propagation delay index matrix; collecting multiple fault response times of the multiple sensing nodes; outputting a fault response time anchor point based on the propagation delay index matrix and the multiple fault response times, performing time series delay calibration on the multiple sensing nodes using the fault response time anchor point, outputting a calibrated time series, and extracting calibrated multi-source perception data under the calibrated time series; using the calibrated multi-source perception data to identify the fault type and output a fault warning signal.

[0005] Preferably, a fault propagation rate model is established based on the multiple spatial position coordinates, and the method also includes: inputting the multiple spatial position coordinates to obtain the signal propagation paths and propagation medium types corresponding to the multiple sensing nodes; collecting historical fault response signals of the multiple sensing nodes under known fault event samples based on the signal propagation paths and propagation medium types; inverting the historical fault response signal samples to set a corresponding preset propagation rate for each sensing node; and constructing a fault propagation rate model based on the relationship between each sensing node and the corresponding preset propagation rate to identify the multiple propagation rates corresponding to the multiple sensing nodes.

[0006] Preferably, the multiple sensing nodes in the multi-source IoT sensing device include at least two of an acoustic sensing device, an image sensing device, a temperature sensing device, a current or voltage sensing device, and a pressure sensing device.

[0007] Preferably, the fault propagation rate model is called to calculate the propagation delay index of each perception node pair, and the method includes: obtaining the first perception node and the second perception node of each perception node pair; obtaining the first spatial position coordinate corresponding to the first perception node, and the second spatial position coordinate corresponding to the second perception node; calling the fault propagation rate model to identify the first propagation rate corresponding to the first perception node, and the second propagation rate corresponding to the second perception node; and outputting the propagation delay index of each perception node pair according to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate, and the second propagation rate.

[0008] Preferably, the propagation delay index of each perception node pair is output according to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate and the second propagation rate. The method includes: outputting the first propagation path length and the second propagation path length according to the first spatial position coordinate and the second spatial position coordinate; obtaining a first propagation response index, the first propagation response index being the ratio of the first propagation path length to the first propagation rate; obtaining a second propagation response index, the second propagation response index being the ratio of the second propagation path length to the second propagation rate; calculating the delay index difference between the first propagation response index and the second propagation response index, and outputting a propagation delay index.

[0009] Preferably, the method for outputting a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times includes: constructing a minimum residual optimization model; using the minimum residual optimization model to calculate the propagation delay indicator matrix and the multiple fault response times, and outputting a fault response time anchor point; wherein the expression of the minimum residual optimization model is: ;in, is the calculated fault response time anchor point, is the optimization variable selected within the candidate fault response time anchor point, is the actual fault response time corresponding to the i-th sensing node, is the propagation path length corresponding to the i-th sensing node, is the propagation rate corresponding to the i-th sensing node, is the propagation response index obtained by theoretical calculation, is the number of sensing nodes.

[0010] Preferably, the time series delay of the plurality of sensing nodes is calibrated based on the fault response time anchor point, and a calibrated time series is output, which is expressed as: ;in, is the response time of the i-th sensor node after time series delay calibration, The default original time point.

[0011] Preferably, the calibrated multi-source perception data is used to identify the fault type and output a fault warning signal. The method includes: using the calibrated multi-source perception data to perform data fusion to obtain fused perception data; extracting the fault feature vector of the fused perception data, identifying the fault type with the fault feature vector, and outputting a fault warning signal corresponding to the fault type.

[0012] In a second aspect, the present invention also provides a device fault warning system under multi-source Internet of Things perception, the system comprising: a node coordinate acquisition module for acquiring multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source Internet of Things sensing device; a propagation rate modeling module for establishing a fault propagation rate model based on the multiple spatial position coordinates; a propagation delay calculation module for combining the multiple sensing nodes to obtain sensing node pairs, calling the fault propagation rate model to calculate the propagation delay index of each sensing node pair, and outputting a propagation delay index matrix; a response time acquisition module for collecting multiple fault response times of the multiple sensing nodes; a delay calibration module for outputting a fault response time anchor point based on the propagation delay index matrix and the multiple fault response times, performing time series delay calibration on the multiple sensing nodes using the fault response time anchor point, outputting a calibration time series, and extracting calibrated multi-source sensing data under the calibration time series; a fault identification and warning module for identifying the fault type using the calibrated multi-source sensing data and outputting a fault warning signal.

[0013] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-mentioned device failure warning method under multi-source Internet of Things perception.

[0014] One or more technical solutions provided in the present invention have at least the following beneficial effects:

[0015] By obtaining multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source IoT sensing device, key spatial dimension information is provided for the subsequent accurate evaluation of fault propagation. A fault propagation rate model is established based on the multiple spatial position coordinates to quantitatively describe the speed and mode of fault propagation between sensing nodes at different spatial locations, laying a theoretical foundation for the subsequent accurate calculation of propagation delay indicators, enabling the entire solution to conduct in-depth analysis of fault propagation from a spatiotemporal perspective. By combining the multiple sensing nodes to obtain sensing node pairs, the fault propagation rate model is called to calculate the propagation delay indicator for each sensing node pair, and a propagation delay indicator matrix is output. This concretizes the abstract fault propagation law into quantifiable indicators, which can clearly reflect the time delay differences caused by fault propagation between different sensing node pairs, providing an accurate quantitative basis for the subsequent time series delay calibration of the sensing nodes. By collecting multiple fault response times of the multiple sensing nodes, the original time information of the sensing nodes under actual fault conditions is obtained. This original time information serves as the benchmark data source for the subsequent time series delay calibration. Based on the propagation delay indicator matrix and the multiple fault response times, a fault response time anchor point is output. The multiple sensing nodes are time-series delayed and calibrated using the fault response time anchor point. A calibrated time series is output, and calibrated multi-source sensing data under the calibrated time series is extracted. Through calibration, the temporal inconsistency of the multi-source sensing data is eliminated, enabling data from different sensing nodes to be compared and fused on a unified time scale, providing a high-quality data foundation for subsequent accurate fault type identification using the calibrated multi-source sensing data. The calibrated multi-source sensing data is used to identify the fault type and output a fault warning signal. The calibrated multi-source sensing data can more accurately reflect the actual operating status of the equipment. Using this data for fault type identification can accurately determine the type of fault and then output a fault warning signal in a timely manner, effectively avoiding equipment damage, production interruption, and other problems caused by untimely fault warnings or false alarms, thereby improving the reliability and stability of equipment operation.

[0016] In summary, the present invention realizes the dynamic modeling of fault propagation between multi-source perception nodes through the collection of spatial position coordinates and the establishment of a propagation rate model, and can accurately characterize the delay characteristics between nodes. Combining the propagation delay index matrix and the actual fault response time, multi-node timing delay calibration is performed based on the fault response time anchor point, which solves the problem of data timing asynchrony caused by the difference in response time of multi-source perception nodes. The calibrated multi-source perception data can more realistically reflect the operating status of the equipment, eliminate the aliasing of fault characteristics, and significantly improve the fusion accuracy and consistency of multi-modal perception data, thereby achieving more accurate and timely fault type identification and early warning, and meeting the high reliability requirements of industrial equipment for health management under complex working conditions.

[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a device failure warning method under multi-source IoT perception provided by an embodiment of the present invention.

[0019] Figure 2 A schematic diagram of a flow chart for calculating the propagation delay index of each sensing node pair in a device fault early warning method under multi-source IoT perception provided by an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of a device failure warning system under multi-source IoT perception provided by an embodiment of the present invention.

[0021] Explanation of reference numerals: node coordinate acquisition module 10 , propagation rate modeling module 20 , propagation delay calculation module 30 , response time acquisition module 40 , delay calibration module 50 , fault identification and warning module 60 . DETAILED DESCRIPTION

[0022] The embodiments of the present invention provide a device fault warning method, system and storage medium under multi-source Internet of Things perception, thereby solving the technical problems in the prior art of insufficient data fusion accuracy and insufficient fault warning accuracy and timeliness due to differences in response time of multi-source perception nodes, and achieving the technical effect of realizing multi-source perception data timing delay calibration, thereby improving fault warning accuracy and timeliness.

[0023] Example 1, as Figure 1 As shown, an embodiment of the present invention provides a device failure warning method under multi-source IoT perception, the method comprising:

[0024] Step S100: Acquire multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source IoT sensing device.

[0025] Furthermore, the multiple sensing nodes in the multi-source IoT sensing device include at least two of an acoustic sensing device, an image sensing device, a temperature sensing device, a current or voltage sensing device, and a pressure sensing device.

[0026] Specifically, a multi-source IoT sensing device refers to a collection of multiple types of sensors deployed within a device or industrial environment, capable of sensing data such as the device's operating status and physical signals. A sensing node refers to the physical sensor unit within a multi-source IoT sensing device that actually performs data collection. In embodiments of the present invention, the multiple sensing nodes within a multi-source IoT sensing device include at least two of the following: acoustic sensing devices, image sensing devices, temperature sensing devices, current or voltage sensing devices, and pressure sensing devices. Acoustic sensing devices, for example, acoustic emission sensors or ultrasonic sensors, use the principle of sound wave propagation to collect acoustic signals generated during device operation. Image sensing devices are sensors based on optical imaging principles, such as industrial cameras or infrared thermal imagers. Temperature sensing devices, such as thermocouples or thermistors, sense temperature signals. Current or voltage sensing devices detect electrical signals using current or voltage sensors. Pressure sensing devices, such as pressure sensors or piezoelectric sensors, detect pressure changes within or around the device. By introducing at least two different types of sensors, multiple aspects of the device's operating status can be covered. For example, by leveraging the complementary nature of acoustic and temperature signals, joint monitoring of mechanical friction anomalies and heat accumulation can be achieved. The integration of multi-modal and multi-type sensing nodes can provide richer physical dimension data, laying a more comprehensive and diversified perception foundation for subsequent fault propagation rate models and data fusion analysis.

[0027] First, the precise spatial coordinates of each multi-source sensing node within the device space are obtained using a known 3D device model or sensor installation data sheet. These spatial coordinates are typically presented as 3D rectangular coordinates, such as a temperature sensor located at (1.2m, 0.5m, 0.3m) or an image sensor located at (0.8m, -0.3m, 0.7m). Using a digital twin or CAD model of the device, the spatial coordinates of the sensing nodes are mapped to the actual device structure, ensuring that all sensing nodes are accurately labeled within the same spatial reference coordinate system. Furthermore, this coordinate data is stored in a location database for subsequent propagation rate modeling and timing delay calibration.

[0028] The spatial position coordinates of multiple sensing nodes obtained through this step can provide an accurate spatial geometric relationship basis for subsequent fault propagation rate modeling, ensuring that the propagation paths and distances between multi-source sensing nodes can be quantified and modeled, thereby improving the spatial consistency and computational accuracy of multi-source sensing data fusion.

[0029] Step S200: establishing a fault propagation rate model according to the multiple spatial position coordinates.

[0030] Specifically, the fault propagation rate model is a mathematical model that combines the spatial location of each sensor node with historical fault signal samples to infer the propagation velocity of each sensor node's sensor response to a fault event. The spatial coordinates of each sensor node are input. The propagation path of the sensor signal between each node is then derived based on the device structure, material information, and sensor type. Furthermore, known historical fault response signal data (such as mechanical shock waves and thermal diffusion characteristics) is obtained from the device. By replaying these historical fault events, the response time of each sensor node to the known fault is recorded. Based on these historical samples, the propagation velocity of the sensor node in different propagation media (such as the speed of sound within metal components and in air) is inferred, thereby constructing a preset propagation velocity for each sensor node. These propagation velocities are then combined with the spatial location relationships to generate a complete propagation rate model.

[0031] Establishing the propagation rate model through this step can truly reflect the signal propagation laws of different sensor nodes in different spatial positions and physical paths, making the subsequent propagation delay indicator calculation more scientific and accurate, and ensuring the time series comparability of the data.

[0032] Step S300: combining the plurality of sensing nodes to obtain sensing node pairs, calling the fault propagation rate model to calculate a propagation delay index of each sensing node pair, and outputting a propagation delay index matrix.

[0033] Specifically, a sensing node pair refers to a combination of any two sensing nodes and is the basic unit used to analyze the signal propagation delay relationship between nodes; the propagation delay index is a quantitative indicator of the delay size of the signal propagating from one sensing node to another; the propagation delay index matrix is a two-dimensional matrix, and each element in the matrix represents the propagation delay index of a node pair.

[0034] All possible sensor node pairs are enumerated and combined. For each sensor node pair, the propagation rate model is applied to obtain the propagation path length and propagation rate for that node pair. The propagation response index is calculated by dividing the propagation path length by the propagation rate. The propagation delay index is then calculated by comparing the propagation response index differences between node pairs. Finally, the propagation delay indexes of all node pairs are summarized in a matrix to form a propagation delay index matrix for subsequent time alignment optimization. For example, consider four sensor nodes A, B, C, and D, and the node pairs AB, AC, AD, and BC. Using the fault propagation rate model, their propagation delays are calculated (e.g., AB is 0.001 seconds, AC is 0.002 seconds, AD is 0.001 seconds, BC is 0.003 seconds, and so on). These delays are then summarized into a 4×4 propagation delay index matrix.

[0035] The propagation delay indicator matrix is obtained through this step, which provides accurate propagation delay constraints for the subsequent time series calibration based on global propagation residual optimization, thereby improving the accuracy of delay calibration.

[0036] Step S400: collecting multiple fault response times of the multiple sensing nodes.

[0037] Specifically, fault response time refers to the timestamp when a sensing node detects a response signal after a device fault event occurs. This is key timing information within the sensing node's data. During device operation, response data from each sensing node is collected in real time, recording the time at which each node perceives the fault signature signal, forming a set of fault response time series. These fault response time series record the latency of each node's perception of the same fault event and serve as crucial raw data for subsequent time calibration.

[0038] Step S500: outputting a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times, performing time series delay calibration on the multiple sensing nodes using the fault response time anchor point, outputting a calibrated time series, and extracting calibrated multi-source sensing data under the calibrated time series.

[0039] Specifically, the fault response time anchor point is the global benchmark time point that best represents the actual fault occurrence time, obtained by optimizing the residual function. Time series delay calibration corrects the time series data of each sensing node based on the anchor point and propagation delay to ensure multi-node data alignment. The corrected result is recorded as the calibrated time series. Calibrated multi-source sensing data is extracted from multi-source sensing data based on the calibrated time series. This data is temporally aligned, allowing for more accurate fusion and analysis.

[0040] First, using a minimum residual optimization model, the propagation delay indicator matrix and the original response time series of the sensing nodes are input to determine a globally optimal fault response time anchor point. Then, based on this fault response time anchor point, the response time of each node is sequentially delayed and corrected according to the propagation response indicator difference, resulting in a set of calibrated time series with consistent timing and minimized error. Finally, based on the calibrated timestamps, the calibrated multi-source sensing data for each node is extracted to ensure temporal consistency for subsequent data fusion analysis.

[0041] This step completes the time series delay calibration of multiple nodes, significantly reduces the timing error in multimodal data fusion, and improves the accuracy of the fusion results.

[0042] Step S600: using the calibrated multi-source sensing data to identify the fault type and output a fault warning signal.

[0043] Specifically, fault type identification involves using fused multimodal features to classify and determine fault types using pattern recognition or machine learning methods. Fault warning signals are warning messages or control instructions issued for detected potential fault types. Based on calibrated multi-source sensor data, feature engineering is first performed to extract key fault feature vectors from the calibrated multi-source sensor data. Next, using machine learning algorithms (such as support vector machines, random forests, and deep neural networks) or rule-based fault feature matching, the multi-source feature vectors are classified or pattern-recognized to determine the type and severity of the fault. Finally, based on the identification results, a corresponding fault warning signal is generated for use by the monitoring center or upper-level control system, enabling real-time warning and remote monitoring of the equipment.

[0044] This step uses multimodal calibration perception data to identify fault types and output reliable warning signals, achieving more accurate and intelligent equipment fault warnings and improving the safety and operation and maintenance efficiency of industrial equipment.

[0045] Furthermore, step S200 includes:

[0046] Step S210: Input the multiple spatial position coordinates and obtain the signal propagation paths and propagation medium types corresponding to the multiple sensing nodes.

[0047] Step S220: collecting historical fault response signals of the plurality of sensing nodes under known fault event samples according to the signal propagation path and propagation medium type.

[0048] Step S230: Invert the historical fault response signal samples and set a corresponding preset propagation rate for each sensing node.

[0049] Step S240: Based on the relationship between each sensing node and the corresponding preset propagation rate, a fault propagation rate model is constructed to identify multiple propagation rates corresponding to the multiple sensing nodes.

[0050] Specifically, a propagation path refers to the possible physical path a signal takes from the fault source (the device being monitored) to each sensing node. The propagation medium type refers to the medium used during signal transmission, such as air, metal, or oil. Using the multiple spatial coordinates corresponding to the sensing nodes obtained in step S100 as input, the propagation path between each sensing node and the potential fault source is determined based on the device's structure, materials, and sensor distribution, combined with a CAD or BIM model. The propagation medium type is identified based on the media involved in the path (e.g., sound wave propagation in air, stress wave propagation in metal, etc.).

[0051] Fault event samples refer to typical fault event data that are known to have occurred and for which the response time of each sensor node is known. Multiple typical fault events (such as mechanical shock and thermal runaway) are extracted from the device's historical operating records. The original signal response sequences and timestamps of each sensor node under these faults are recorded. During collection, the response signals of each node are ensured to be reproducible and cover a variety of propagation paths and media combinations to enhance the generalization capability of the propagation rate model.

[0052] Inversion involves inferring the propagation rate based on response time differences, spatial location, and path information. The preset propagation rate is the propagation rate set for each sensing node during the model initialization phase. Using methods such as least squares or genetic algorithms, the propagation rate for each node pair is inferred using the response time differences, propagation path lengths, and propagation medium types of each node to a known fault. The propagation rates of each node under multiple historical samples are then averaged or weighted to determine the preset propagation rate for each sensing node.

[0053] Based on the preset propagation rates and the spatial distribution of nodes, a rate model is constructed that incorporates multidimensional information about propagation rate, path length, and the propagation medium. This model can use a matrix form, a graph neural network, or a propagation delay expression based on physical formulas. It can quickly query the propagation rate of each sensing node pair, which is then used to generate the propagation delay indicator matrix.

[0054] Further, such as Figure 2 As shown, step S300 includes:

[0055] Step S310: Obtain the first sensory node and the second sensory node of each sensory node pair.

[0056] Step S320: Obtain the first spatial position coordinates corresponding to the first sensing node and the second spatial position coordinates corresponding to the second sensing node.

[0057] Step S330: calling the fault propagation rate model to identify a first propagation rate corresponding to the first sensing node and a second propagation rate corresponding to the second sensing node.

[0058] Step S340: Output the propagation delay indicator of each sensing node pair according to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate, and the second propagation rate.

[0059] Specifically, all sensor nodes are paired up to form several sensor node pairs. Each node pair consists of a first sensor node and a second sensor node, and their numbers or IDs are recorded separately. The three-dimensional spatial coordinates of the first and second sensor nodes are retrieved from a database storing multiple spatial coordinates and recorded as the first and second spatial coordinates.

[0060] Using the previously established fault propagation rate model, the first and second sensing nodes are used as inputs. Through table lookup, extrapolation, and model calculation, the first propagation rate corresponding to the first sensing node and the second propagation rate corresponding to the second sensing node are obtained. Based on the acquired spatial coordinates and the identified propagation rates, the spatial distances from the first and second sensing nodes to the fault source are first calculated. Next, the spatial distance from the first spatial coordinate to the fault point (the first propagation path length) is divided by the first propagation rate, and the spatial distance from the second spatial coordinate to the fault point (the second propagation path length) is divided by the second propagation rate, respectively, to determine the respective propagation response times. The difference between the two values is the propagation delay index. Finally, the propagation delay indexes for all node pairs are organized into a matrix to obtain the propagation delay index matrix, which facilitates the global optimization of subsequent time anchor points.

[0061] Furthermore, step S340 includes:

[0062] Step S341: outputting a first propagation path length and a second propagation path length according to the first spatial position coordinates and the second spatial position coordinates.

[0063] Step S342: Obtain a first propagation response indicator, where the first propagation response indicator is a ratio of the first propagation path length to the first propagation rate.

[0064] Step S343: Obtain a second propagation response indicator, where the second propagation response indicator is a ratio of the second propagation path length to the second propagation rate.

[0065] Step S344: Calculate the delay indicator difference between the first propagation response indicator and the second propagation response indicator, and output a propagation delay indicator.

[0066] Specifically, the propagation response index (PRI) is the product of dividing the signal propagation path length by the propagation rate, representing the propagation time from the node to the fault point. Using the three-dimensional Euclidean distance formula, the straight-line distances between the first and second sensing nodes and the fault source are calculated, respectively, to obtain the first and second propagation path lengths. Based on the known first propagation path length and first propagation rate, the ratio of the first propagation path length to the first propagation rate is calculated, denoted as the first propagation response index. Similarly, the ratio of the second propagation path length to the second propagation rate is calculated, denoted as the second propagation response index. The absolute value of the difference between the two propagation response indices is used to determine the propagation delay index for the first and second sensing nodes. A larger value for the propagation delay index indicates a more significant difference in the responses of the first and second sensing nodes to the same fault event. Using this calculation method, the propagation delay indexes for all pairs of sensing nodes are determined and combined into a propagation delay index matrix to intuitively reflect the response differences between multiple sensing nodes.

[0067] Furthermore, step S500 includes:

[0068] Step S510: constructing a minimum residual optimization model.

[0069] Step S520: Calculate the propagation delay indicator matrix and the multiple fault response times using the minimum residual optimization model, and output the fault response time anchor point. The expression of the minimum residual optimization model is: ;in, is the calculated fault response time anchor point, is the optimization variable selected within the candidate fault response time anchor point, is the actual fault response time corresponding to the i-th sensing node, is the propagation path length corresponding to the i-th sensing node, is the propagation rate corresponding to the i-th sensing node, is the propagation response index obtained by theoretical calculation, is the number of sensing nodes.

[0070] Specifically, the minimum residual optimization model is a mathematical optimization model that finds the optimal fault response time anchor point by minimizing the sum of squared differences between the theoretical propagation response index and the actual fault response time. The fault response time anchor point refers to a globally unified theoretical fault response starting time point, which serves as a reference for calibrating the time series of multiple sensing nodes. The candidate fault response time anchor point refers to the range of values of the optimization variable T in the minimum residual optimization model, that is, the multiple possible fault response time starting points attempted during the search process. This range can be set based on the distribution of the actual response times of multiple sensing nodes, equipment operating conditions, or historical experience.

[0071] First, based on the propagation delay index matrix and the actual fault response time of each sensing node, a minimum residual optimization model is built. The objective function of this model is to minimize the sum of squared errors between the theoretical propagation delay index and the actual response time of each sensing node. The formal expression is as follows: ,in, is the calculated fault response time anchor point, is the optimization variable selected within the candidate fault response time anchor point, is the actual fault response time corresponding to the i-th sensing node, is the propagation path length corresponding to the i-th sensing node, is the propagation rate corresponding to the i-th sensing node, is the propagation response index obtained by theoretical calculation, is the number of sensing nodes.

[0072] Call the minimum residual optimization model constructed above and input all perception nodes ( The actual response time of sensor nodes (n is a positive integer) ), propagation path length ( ) and propagation rate ( ), and the propagation delay indicator matrix. The mean range of all theoretical propagation delay indicators is used as the center of the candidate fault response time anchor point. Then, combined with the actual working conditions of the multi-source IoT perception system (such as the response delay period after the device is started or the sensor is stable), a certain time range is extended forward and backward (such as ±10ms or ±5%) to form a set of candidate fault response time anchor points. The minimum residual optimization model iteratively searches within the range of candidate fault response time anchor points (T) until the sum of squared errors is minimized and the optimal output is obtained. The value serves as the global fault response time anchor point for the entire perception system. This iterative search process can be implemented using optimization methods such as gradient descent and least squares fitting, or it can be automatically solved using existing numerical optimization libraries to ultimately obtain a precise fault response time anchor point.

[0073] Furthermore, the time series delay of the plurality of sensing nodes is calibrated based on the fault response time anchor point, and the calibrated time series is output, which is expressed as follows: ;in, is the response time of the i-th sensor node after time series delay calibration, The default original time point.

[0074] Specifically, the fault response time anchor point obtained by minimum residual optimization is , for the original response time series collected by each sensing node, based on the propagation path length ( ) and propagation rate ( ), combined with the preset original time point ( ), perform time series delay calibration to eliminate the time delay differences of multi-source perception data, align the response times of each perception node under the same global time reference, and obtain the calibrated response time of each perception node under the global time reference, that is, the calibrated time series. The specific calibration expression is as follows: ;in, is the response time of the i-th sensor node after time series delay calibration, The default original time point, usually set to 0.

[0075] The core of this calibration step is to quantitatively compensate for the time delay of each sensing node, so that the response time series of each node are synchronized within the framework of the theoretical propagation model, avoiding the time drift problem caused by different node distribution locations or differences in propagation media, and thus providing a temporal basis for subsequent multimodal data fusion and fault type identification, so as to achieve more efficient and reliable equipment fault warning.

[0076] Furthermore, step S600 includes:

[0077] Step S610: performing data fusion using the calibrated multi-source perception data to obtain fused perception data.

[0078] Step S620: extracting the fault feature vector of the fused sensing data, identifying the fault type with the fault feature vector, and outputting a fault warning signal corresponding to the fault type.

[0079] Specifically, data fusion refers to the integration of multi-source data from different sensing nodes according to a specific fusion strategy to form complete, consistent, and minimally redundant global perception information. A fault feature vector is a set of features extracted from the fused multimodal perception data that characterizes the fault mode. These features can be represented using statistical features, time-frequency domain features, or deep features (such as convolutional feature vectors).

[0080] First, calibrated multi-source perception data is acquired from each sensing node and data fusion is performed under the premise of time alignment. Data fusion can be performed using methods such as weighted averaging, multimodal feature concatenation, Bayesian reasoning, or multimodal fusion using deep neural networks. For example, tools such as TensorFlow and PyTorch can be used to perform tensor concatenation or self-attention integration to generate fused perception data. Subsequently, feature vectors are extracted from the fused perception data to distinguish fault types. These features can include statistical indicators of time series (such as mean, standard deviation, and peak value), spectral energy distribution (such as the amplitude spectrum after Fourier transform), or deep features extracted using convolutional neural networks. Finally, based on known training samples or expert experience, classifiers such as support vector machines, random forests, and convolutional neural networks are used to perform pattern recognition on the fault feature vectors. The fault type identification result is output and a corresponding fault warning signal is generated. To ensure real-time and scalability, this process is typically performed on edge computing gateways or cloud servers to ensure that the fused perception data can be quickly converted into accurate warning results.

[0081] Taking wind turbine equipment as an example, multimodal data collected by multiple sensing nodes (including acoustic sensors, temperature sensors, and current sensors) installed on the turbine are used. After time series delay calibration, calibrated acoustic signals, temperature change curves, and current fluctuation sequences are obtained. Next, data fusion is performed on the calibrated multimodal sensing data: the acoustic signals are converted into time-frequency spectrograms using a short-time Fourier transform. The temperature and current time series signals are normalized and then concatenated into the channel dimensions of the acoustic time-frequency spectrogram to form a multi-channel fused sensing data tensor. This fused sensing data tensor is then input into a fault diagnosis model based on a convolutional neural network. The convolutional layer uses multiple convolution kernels to extract local features of the fused sensing data in the spatial and time-frequency domains. The pooling layer performs feature compression to remove noise and redundant information. Subsequently, the fully connected layer maps and fuses high-level semantic features. Finally, the softmax layer outputs the probability distribution of each fault type. For example, the fault diagnosis model was trained to distinguish three fault types: bearing wear, gear defects, and normal conditions. When the model infers a set of calibrated fusion data, the softmax layer outputs the highest probability of the bearing wear category, indicating that the fault type matched by the current multimodal perception data is bearing wear. Finally, based on the output of the fault diagnosis model, a corresponding fault warning signal is generated and pushed to the operation and maintenance personnel or remote monitoring center, realizing intelligent early warning.

[0082] By fusing and calibrating multi-source sensor data, we can fully leverage the complementary advantages of multimodal sensing, overcome the limitations of a single sensing method, and effectively improve the accuracy and robustness of fault identification. Ultimately, fault type identification and warning signal output are performed based on feature vectors extracted from the fused sensor data, significantly improving the comprehensive perception capability and intelligent warning level of the equipment's operating status, and achieving closed-loop control from multi-source data acquisition to accurate warnings.

[0083] In summary, the device fault warning method under multi-source IoT perception provided by the embodiment of the present invention has the following beneficial effects:

[0084] An embodiment of the present invention provides a device fault warning method under multi-source Internet of Things perception. First, based on the spatial position coordinates corresponding to multiple sensing nodes in the multi-source Internet of Things perception device, a fault propagation rate model that can reflect the response differences of multiple nodes is established; by combining sensing node pairs and calling the fault propagation rate model, the propagation delay index matrix of each sensing node pair is derived, effectively quantifying the delay deviation caused by the spatial topology and propagation rate differences; further, the actual fault response time of each sensing node is collected, and the fault response time anchor point is derived through the minimum residual optimization model. Combined with the propagation delay index and the preset time point, the time series delay calibration formula is used to perform time series correction on the data of all sensing nodes; finally, based on the calibrated multi-source perception data, data fusion and feature extraction are used to derive the fault feature vector and identify the fault type, thereby realizing high-precision and low-latency fault warning by integrating multi-dimensional perception data.

[0085] In general, the embodiment of the present invention realizes the dynamic modeling of fault propagation between multi-source perception nodes through the collection of spatial position coordinates and the establishment of a propagation rate model, and can accurately characterize the delay characteristics between nodes. Combining the propagation delay index matrix and the actual fault response time, multi-node timing delay calibration is performed based on the fault response time anchor point, which solves the problem of data timing asynchrony caused by the difference in response time of multi-source perception nodes. The calibrated multi-source perception data can more realistically reflect the operating status of the equipment, eliminate the aliasing of fault characteristics, and significantly improve the fusion accuracy and consistency of multi-modal perception data, thereby achieving more accurate and timely fault type identification and warning, and meeting the high reliability requirements of industrial equipment for health management under complex working conditions.

[0086] Example 2, as Figure 3 As shown, based on the same inventive concept as the aforementioned embodiment 1, an embodiment of the present invention provides a device failure warning system under multi-source IoT perception, the system comprising:

[0087] The node coordinate acquisition module 10 is used to obtain multiple spatial position coordinates corresponding to multiple sensing nodes in the multi-source IoT sensing device.

[0088] The propagation rate modeling module 20 is configured to establish a fault propagation rate model according to the plurality of spatial position coordinates.

[0089] The propagation delay calculation module 30 is configured to combine the plurality of sensing nodes to obtain sensing node pairs, call the fault propagation rate model to calculate the propagation delay index of each sensing node pair, and output a propagation delay index matrix.

[0090] The response time collection module 40 is configured to collect multiple fault response times of the multiple sensing nodes.

[0091] The delay calibration module 50 is used to output a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times, perform time series delay calibration on the multiple sensing nodes using the fault response time anchor point, output a calibrated time series, and extract calibrated multi-source sensing data under the calibrated time series.

[0092] The fault identification and warning module 60 is configured to use the calibrated multi-source sensing data to identify the fault type and output a fault warning signal.

[0093] Furthermore, the multiple sensing nodes in the multi-source IoT sensing device include at least two of an acoustic sensing device, an image sensing device, a temperature sensing device, a current or voltage sensing device, and a pressure sensing device.

[0094] Furthermore, the propagation rate modeling module 20 is further configured to perform the following steps:

[0095] The multiple spatial position coordinates are input to obtain the signal propagation paths and propagation medium types corresponding to the multiple sensing nodes; based on the signal propagation paths and propagation medium types, historical fault response signals of the multiple sensing nodes under known fault event samples are collected; the historical fault response signal samples are inverted to set a corresponding preset propagation rate for each sensing node; and based on the relationship between each sensing node and the corresponding preset propagation rate, a fault propagation rate model is constructed to identify the multiple propagation rates corresponding to the multiple sensing nodes.

[0096] Furthermore, the propagation delay calculation module 30 is further configured to perform the following steps:

[0097] Obtain a first sensing node and a second sensing node of each sensing node pair; obtain a first spatial position coordinate corresponding to the first sensing node, and a second spatial position coordinate corresponding to the second sensing node; call the fault propagation rate model to identify a first propagation rate corresponding to the first sensing node, and a second propagation rate corresponding to the second sensing node; output a propagation delay indicator of each sensing node pair according to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate, and the second propagation rate.

[0098] Furthermore, the propagation delay calculation module 30 is further configured to perform the following steps:

[0099] According to the first spatial position coordinates and the second spatial position coordinates, the first propagation path length and the second propagation path length are output; a first propagation response index is obtained, where the first propagation response index is the ratio of the first propagation path length to the first propagation rate; a second propagation response index is obtained, where the second propagation response index is the ratio of the second propagation path length to the second propagation rate; a delay index difference between the first propagation response index and the second propagation response index is calculated, and a propagation delay index is output.

[0100] Furthermore, the delay calibration module 50 is further configured to perform the following steps:

[0101] Constructing a minimum residual optimization model; using the minimum residual optimization model to calculate the propagation delay indicator matrix and the multiple fault response times, and outputting the fault response time anchor point; wherein the expression of the minimum residual optimization model is: ;in, is the calculated fault response time anchor point, is the optimization variable selected within the candidate fault response time anchor point, is the actual fault response time corresponding to the i-th sensing node, is the propagation path length corresponding to the i-th sensing node, is the propagation rate corresponding to the i-th sensing node, is the propagation response index obtained by theoretical calculation, is the number of sensing nodes.

[0102] Furthermore, the fault identification and warning module 60 performs time series delay calibration on the multiple sensing nodes based on the fault response time anchor point, and outputs a calibrated time series, which is expressed as: ;in, is the response time of the i-th sensor node after time series delay calibration, The default original time point.

[0103] Furthermore, the fault identification and warning module 60 is further configured to perform the following steps:

[0104] The calibrated multi-source perception data is used to perform data fusion to obtain fused perception data; a fault feature vector of the fused perception data is extracted, a fault type is identified using the fault feature vector, and a fault warning signal corresponding to the fault type is output.

[0105] Through the above detailed description of the device fault warning method under multi-source Internet of Things perception in this specification, those skilled in the art can clearly understand the device fault warning system under multi-source Internet of Things perception in this embodiment. For the system disclosed in Example 2, since it corresponds to the method disclosed in Example 1 and has corresponding functional modules and beneficial effects, the relevant parts can be referred to the method part description.

[0106] Example 3. Based on the same inventive concept of the device fault warning method under multi-source Internet of Things perception in the aforementioned Example 1, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the various steps of the above-mentioned device fault warning method embodiment under multi-source Internet of Things perception are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be repeated here.

[0107] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A device failure early warning method based on multi-source IoT perception is characterized by: The method comprises: Obtain multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source IoT sensing device; establishing a fault propagation rate model based on the plurality of spatial position coordinates; Combining the plurality of sensing nodes to obtain sensing node pairs, calling the fault propagation rate model to calculate a propagation delay index of each sensing node pair, and outputting a propagation delay index matrix; Collecting multiple fault response times of the multiple sensing nodes; Outputting a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times, performing time series delay calibration on the multiple sensing nodes using the fault response time anchor point, outputting a calibrated time series, and extracting calibrated multi-source sensing data under the calibrated time series; The calibrated multi-source sensing data is used to identify the fault type and output a fault warning signal.

2. The device failure early warning method under multi-source IoT perception as claimed in claim 1, characterized in that: Establishing a fault propagation rate model based on the multiple spatial position coordinates, the method further includes: Input the multiple spatial position coordinates to obtain signal propagation paths and propagation medium types corresponding to the multiple sensing nodes; Collecting historical fault response signals of the plurality of sensing nodes under known fault event samples according to the signal propagation path and propagation medium type; Inverting the historical fault response signal samples and setting a corresponding preset propagation rate for each sensing node; Based on the relationship between each sensing node and the corresponding preset propagation rate, a fault propagation rate model is constructed to identify multiple propagation rates corresponding to the multiple sensing nodes.

3. The device failure early warning method under multi-source IoT perception as claimed in claim 1, characterized in that: The multiple sensing nodes in the multi-source IoT sensing device include at least two of an acoustic sensing device, an image sensing device, a temperature sensing device, a current or voltage sensing device, and a pressure sensing device.

4. The device failure early warning method under multi-source IoT perception as claimed in claim 1, characterized in that: The fault propagation rate model is called to calculate the propagation delay index of each sensing node pair, and the method includes: Obtain a first sensory node and a second sensory node of each sensory node pair; Obtaining a first spatial position coordinate corresponding to the first sensing node and a second spatial position coordinate corresponding to the second sensing node; Invoking the fault propagation rate model to identify a first propagation rate corresponding to the first sensing node and a second propagation rate corresponding to the second sensing node; According to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate, and the second propagation rate, a propagation delay indicator of each sensing node pair is output.

5. The device failure warning method under multi-source IoT perception as claimed in claim 4 is characterized in that: Outputting a propagation delay indicator of each sensing node pair according to the first spatial position coordinate, the second spatial position coordinate, the first propagation rate, and the second propagation rate, the method comprising: Outputting a first propagation path length and a second propagation path length according to the first spatial position coordinates and the second spatial position coordinates; Obtaining a first propagation response indicator, where the first propagation response indicator is a ratio of the first propagation path length to the first propagation rate; Obtaining a second propagation response indicator, where the second propagation response indicator is a ratio of the second propagation path length to the second propagation rate; A delay indicator difference between the first propagation response indicator and the second propagation response indicator is calculated, and a propagation delay indicator is output.

6. The device failure early warning method under multi-source IoT perception as claimed in claim 1, characterized in that: Outputting a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times, the method includes: Construct a minimum residual optimization model; Calculating the propagation delay indicator matrix and the multiple fault response times using the minimum residual optimization model, and outputting a fault response time anchor point; Among them, the expression of the minimum residual optimization model is: ； in, is the calculated fault response time anchor point, is the optimization variable selected within the candidate fault response time anchor point, is the actual fault response time corresponding to the i-th sensing node, is the propagation path length corresponding to the i-th sensing node, is the propagation rate corresponding to the i-th sensing node, is the propagation response index obtained by theoretical calculation, is the number of sensing nodes.

7. The device failure warning method under multi-source IoT perception as claimed in claim 6, characterized in that: The time series delay of the multiple sensing nodes is calibrated using the fault response time anchor point, and a calibrated time series is output, which is expressed as: ； in, is the response time of the i-th sensor node after time series delay calibration, The default original time point.

8. The device failure early warning method under multi-source IoT perception as claimed in claim 1, characterized in that: The method of using the calibrated multi-source sensing data to identify the fault type and output a fault warning signal includes: Performing data fusion using the calibrated multi-source perception data to obtain fused perception data; Extract the fault feature vector of the fused perception data, identify the fault type using the fault feature vector, and output a fault warning signal corresponding to the fault type.

9. The equipment failure warning system under multi-source IoT perception is characterized by: The system is used to execute the device failure early warning method under multi-source Internet of Things perception according to any one of claims 1 to 8, comprising: A node coordinate acquisition module is used to obtain multiple spatial position coordinates corresponding to multiple sensing nodes in a multi-source IoT sensing device; a propagation rate modeling module, configured to establish a fault propagation rate model according to the plurality of spatial position coordinates; a propagation delay calculation module, configured to combine the plurality of sensing nodes to obtain sensing node pairs, invoke the fault propagation rate model to calculate a propagation delay index for each sensing node pair, and output a propagation delay index matrix; A response time collection module, configured to collect multiple fault response times of the multiple sensing nodes; a delay calibration module, configured to output a fault response time anchor point based on the propagation delay indicator matrix and the multiple fault response times, perform time series delay calibration on the multiple sensing nodes using the fault response time anchor point, output a calibrated time series, and extract calibrated multi-source sensing data under the calibrated time series; The fault identification and warning module is used to use the calibrated multi-source sensing data to identify the fault type and output a fault warning signal.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the device failure warning method under multi-source Internet of Things perception as described in any one of claims 1 to 8 are implemented.

Images