Information sensing control system and method based on industrial internet of things cloud service

Abstract

The application discloses an information sensing control system and method based on industrial internet of things cloud service, relates to the technical field of industrial internet of things and cloud computing, and is specifically directed to an industrial application scene, and realizes the system and method of intelligent regulation and control of industrial equipment, wherein a data acquisition module collects equipment operation data, a data processing module carries out cleaning analysis and generates a report, a decision algorithm module generates a control instruction based on a multi-objective optimization algorithm, and an execution module adjusts the equipment operation state, so that the system has the functions of abnormality detection and real-time monitoring, the scheduling efficiency is improved through dynamic task allocation, intelligent equipment access realizes plug and play, data hierarchical processing relieves network congestion, predictive maintenance is realized based on real-time data, the intelligent level of the system is significantly improved, and the information sensing method suitable for industrial applications is provided.

Description

Technical Field

[0001] This invention relates to the fields of industrial Internet of Things (IIoT) and cloud computing technology, specifically to an information perception control system and method based on IIoT cloud services. Background Technology

[0002] In the process of using the Internet of Things (IoT) in industry to achieve device interconnection, data exchange, and remote control, traditional technical solutions face many technical bottlenecks that urgently need to be addressed.

[0003] Existing systems generally employ static task allocation strategies, failing to dynamically schedule tasks based on the real-time load status of edge computing nodes and cloud servers. This results in low utilization of computing resources and increased response latency. In the data acquisition phase, conventional polling mechanisms fail to incorporate key parameters such as command transmission latency and device response time into the optimization of acquisition strategies, making it difficult to meet the real-time requirements of industrial scenarios. Device access management still heavily relies on manual configuration, lacking an intelligent access mechanism that integrates automatic device identification, identity authentication, and security verification, severely impacting the rapid deployment and collaborative operation efficiency of heterogeneous devices. At the data processing level, the lack of effective data classification and grading strategies means that all acquired data is directly uploaded to the cloud processing center in its raw form, wasting network bandwidth resources and easily causing delays in critical data transmission. In the field of equipment health monitoring, the single threshold alarm mechanism can only reflect the local state of equipment operation, failing to integrate multi-dimensional operating parameters such as vibration, temperature, and current and conduct comprehensive evaluation through machine learning algorithms, resulting in significantly insufficient accuracy in equipment life prediction. The current planned maintenance model uses fixed maintenance cycles, failing to build predictive maintenance models based on real-time equipment operating status data, leading to unreasonable allocation of maintenance resources and increased risk of unplanned equipment downtime.

[0004] These technical deficiencies severely restrict the improvement of the application efficiency of industrial IoT systems in the manufacturing field. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an information sensing control system and method based on industrial IoT cloud services. The technical problems to be solved are: the use of static task allocation strategies makes it impossible to dynamically schedule tasks according to computing load; the data acquisition stage relies on conventional polling mechanisms and does not consider command transmission latency and equipment response time parameters; the data processing stage lacks a hierarchical strategy, resulting in insufficient accuracy in equipment lifespan prediction; and the maintenance mode uses a fixed-cycle plan. This invention provides an information sensing control system and method based on industrial IoT cloud services, solving the above problems.

[0006] The objective of this invention is mainly achieved through the following technical solutions:

[0007] The information perception and control system based on industrial IoT cloud services includes: a data acquisition module, a data processing module, a decision algorithm module, and an execution module.

[0008] in,

[0009] The data acquisition module is used to collect raw data on the operating status of industrial equipment and transmit the raw data to the data processing module via the cloud.

[0010] The data processing module performs outlier detection and hierarchical processing on the received raw data, generates a data report on the operation of industrial equipment on the production line, and sends the data report to the decision algorithm module.

[0011] The decision algorithm module generates corresponding control commands based on the received data reports using a preset algorithm, and then sends the control commands to the execution module.

[0012] The execution module receives control commands and converts them into physical control signals that can be executed by the industrial equipment, thereby adjusting the operating status of the industrial equipment.

[0013] To further optimize the above technical solution, the data acquisition module collects the current waveform and vibration spectrum of the industrial equipment during operation as raw data.

[0014] The data acquisition module uses a multimodal sensor array to collect equipment operation data. The multimodal sensor array includes vibration sensors and current sensors.

[0015] To further optimize the above technical solution, the data processing module uses the sliding window method to standardize the original data to obtain time-series data, and uses the Grubbs criterion to detect and remove outliers. A hierarchical processing mechanism is established to predict the lifespan of industrial equipment based on the data after outlier removal, and hierarchical processing is performed based on the predicted lifespan of industrial equipment.

[0016] To further optimize the above technical solution, outlier detection adopts the Grubbs criterion, specifically including:

[0017] Time series data As sample data, denoted as Where n is the number of sample data, This is the nth sample data.

[0018] For any data Calculate its Grubbs statistic. The calculation formula is: .

[0019] Where i is any natural number, It is the sample mean, calculated using the following formula: s is the sample standard deviation, calculated using the formula: .

[0020] Based on the number of sample data n and the set significance level α, the Grubbs critical value is obtained by looking up a table. .

[0021] Compare the Grubbs statistic for each data point With critical value ,like Then determine This is an outlier.

[0022] Remove all data that is deemed abnormal.

[0023] Recalculate the mean and standard deviation for the data after removing outliers, and repeat the above steps until all data are processed. .

[0024] To further optimize the above technical solutions, industrial equipment life prediction specifically includes:

[0025] The time series data after outlier removal is converted into a frequency domain signal using a Fast Fourier Transform to extract frequency components.

[0026] Then, the frequency components of the frequency domain signal are captured by short-time Fourier transform to extract time-varying features and calculate power spectral density to obtain the time-varying features of power distribution protrusion faults.

[0027] Calculate the contribution of each time-varying feature to the equipment life prediction, and select the feature with the most predictive power as the frequency domain feature.

[0028] Based on frequency domain characteristics, a convolutional neural network is used to build an industrial equipment life prediction model in the cloud.

[0029] Process newly acquired data in real time to predict remaining lifespan.

[0030] To further optimize the above technical solution, the Fast Fourier Transform converts time-series data into frequency domain signals, specifically including: converting time-series data... Convert to frequency domain signal The formula is expressed as .

[0031] Where t represents the time point of the sampling point in the time series data, It is a time-domain signal, where Ts is the total number of samples and k is the frequency parameter. The frequency domain amplitude is the signal strength corresponding to the frequency component k.

[0032] Extracting time-varying features specifically includes:

[0033] The short-time Fourier transform is used to extract continuous time-varying features. The calculation formula is as follows:

[0034]

[0035] in, It is time-series data; It is a time window function used to extract data from time windows. The local signal segment centered on the Hanning window is used. Tn is the signal length; It is the frequency domain basis function of the Fourier transform, which converts the time-domain signal into a component with frequency k; is the time parameter, i.e., the center position of the window, and k is the frequency parameter.

[0036] Instantaneous energy can be obtained at this time. and spectral centroid As a time-varying feature.

[0037] Instantaneous energy and spectral centroid The calculation formula is: , ;

[0038] The power spectral density, i.e., the power distribution at frequency k of the signal in the frequency domain, is calculated using the following formula:

[0039]

[0040] Constructing the time-varying feature matrix

[0041] Where i and j are arbitrary natural numbers, and q is the type of fault. It is the time-varying characteristic matrix of the q-th type of fault in industrial equipment. It's the first frequency magnitude. It is the magnitude of the i-th frequency. The first one The central time window The j-th The central time window

[0042] The mutual information method is used to calculate the contribution of time-varying features. The calculation formula is as follows:

[0043]

[0044] Where y is any natural number, It is the total lifespan of industrial equipment. It is a time-varying characteristic matrix The joint probability that the lifetime Y takes the value f and the lifetime Y takes the value y; yes The probability of taking the value f; It is the probability that the lifespan Y takes the value y.

[0045] Take contribution The time-varying characteristics of the two or more highest faults are used as frequency domain feature B. .

[0046] To further optimize the above technical solution, the tiered processing mechanism is based on the remaining life predicted by the industrial equipment life prediction model. Calculate the health status of industrial equipment The calculation formula is: Y represents the total lifespan of the industrial equipment.

[0047] Set three-level processing thresholds, set thresholds ;

[0048] when Interlock shutdown, when Normal operation at times, when and Maintenance alerts were sent to cloud users.

[0049] when At this time, it is necessary to adjust the equipment load. Normal operation at times, when and Generates running data reports every hour.

[0050] To further optimize the above technical solution, the decision algorithm module includes a multi-objective optimization algorithm. The multi-objective optimization algorithm establishes a vector space including energy consumption parameters and efficiency parameters, solves the extreme points of the Lagrange function under multi-objective constraints as the optimal solution, and generates control commands based on the optimal solution.

[0051] To further optimize the above technical solution, a monitoring module is also included. The monitoring module monitors the execution of control commands by the execution module in real time as the monitoring result, tracks command execution deviations, and feeds the monitoring results back to the data processing module.

[0052] The execution status is obtained by collecting displacement signals from displacement sensors and torque feedback signals from motor outputs generated by the execution module in accordance with control commands in real time, and constructing a time series dataset of the control command execution status as the monitoring result.

[0053] An information sensing and control method based on industrial IoT cloud services, applied to an information sensing and control system based on industrial IoT cloud services, includes the following steps:

[0054] Step S1: Collect raw data of industrial equipment operation through the data acquisition module.

[0055] Step S2: Transmit the raw data to the data processing module for data cleaning, processing and analysis to generate a data report.

[0056] Step S3: Input the data report into the decision algorithm module and use the preset algorithm to generate control commands.

[0057] Step S4: The execution module receives control commands, executes the commands according to the control commands, and regulates the operating status of the corresponding industrial equipment.

[0058] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the information sensing control system and method based on industrial Internet of Things cloud services.

[0059] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods in the information sensing control system and method based on industrial Internet of Things cloud services.

[0060] A computer program product includes a computer program that, when executed by a processor, implements the steps of any one of the methods in the information sensing control system and method based on industrial Internet of Things cloud services.

[0061] In summary, the present invention has the following advantages compared with the prior art:

[0062] Through the collaborative work of data acquisition, data processing, decision-making algorithm, and execution modules, this invention achieves dynamic acquisition, intelligent processing, and precise control of industrial equipment operating data. It solves the problems of low resource utilization, high response latency, and insufficient maintenance efficiency in traditional systems. The multimodal sensor array and cloud transmission mechanism ensure the real-time performance and integrity of data acquisition; anomaly detection and hierarchical processing significantly improve data quality; multi-objective optimization algorithms enhance the accuracy of control command generation; and programmable multi-protocol interfaces solve the problem of heterogeneous device access. The system organically integrates sensing, analysis, decision-making, and execution through a cloud-based collaborative architecture, effectively improving the control response speed and device compatibility of industrial IoT systems. It offers advantages such as improved data processing efficiency, optimized resource scheduling, enhanced equipment status monitoring accuracy, and predictive maintenance. Attached Figure Description

[0063] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0064] Figure 1 A flowchart of an information sensing and control method based on cloud-based industrial IoT services;

[0065] Figure 2 This is a flowchart illustrating the steps of an information sensing and control method based on cloud-based industrial IoT services. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for illustrative purposes only and are not intended to limit the invention. The embodiments of this application will now be described with reference to the accompanying drawings.

[0067] Traditional solutions employ static task allocation strategies, which cannot dynamically schedule tasks based on computing load, resulting in low resource utilization and increased response latency. The data acquisition stage relies on conventional polling mechanisms, failing to consider command transmission latency and device response time parameters, making it difficult to meet real-time requirements. The data processing stage lacks a hierarchical strategy, with raw data being directly uploaded, leading to bandwidth waste and delays in critical data. Health monitoring uses single-threshold alarms, failing to integrate multi-dimensional parameters for comprehensive evaluation, resulting in insufficient accuracy in device lifespan prediction. The maintenance mode uses fixed-cycle plans, failing to build predictive models based on real-time status, thus posing a risk of unplanned downtime.

[0068] In view of this, the present invention provides the following embodiments to solve the above-mentioned technical problems.

[0069] In Example 1:

[0070] Reference Figure 1 , Figure 2 This is the first embodiment of the present invention. This embodiment provides an information perception control system and method based on industrial Internet of Things cloud services, which has the advantages of improving data processing efficiency, enhancing equipment status monitoring accuracy, and realizing predictive maintenance.

[0071] like Figure 1 As shown, the information perception and control system based on industrial IoT cloud services includes a data acquisition module, a data processing module, a decision algorithm module, and an execution module.

[0072] The data acquisition module is used to collect raw data on the operating status of industrial equipment and transmit the raw data to the data processing module via the cloud.

[0073] The data processing module performs outlier detection and hierarchical processing on the received raw data to generate a data report on the operation of industrial equipment on the production line, and sends the data report to the decision algorithm module.

[0074] The decision algorithm module generates corresponding control commands based on the received data reports using a preset algorithm, and sends the control commands to the execution module.

[0075] The execution module receives control commands and converts them into physical control signals that can be executed by industrial equipment.

[0076] like Figure 2 As shown, the information sensing and control method based on industrial IoT cloud services mainly includes the following steps:

[0077] Step S1: Collect raw data on the operation of industrial equipment through the data acquisition module;

[0078] Step S2: Transmit the raw data to the data processing module for data cleaning, processing and analysis to generate a data report;

[0079] Step S3: Input the data report into the decision algorithm module and generate control commands using a preset algorithm;

[0080] Step S4: The execution module receives the control command, executes the command according to the control command, and regulates the operating status of the corresponding industrial equipment.

[0081] To address existing problems, this embodiment uses a data acquisition module to collect raw data on the operating status of industrial equipment and transmit it to the cloud. A data processing module performs outlier detection and hierarchical processing to generate data reports. A decision-making algorithm module generates control commands based on the data reports, and an execution module converts these control commands into physical control signals. Multiple actuator interfaces are configured to support different types of equipment access, constructing a complete closed-loop management system. Firstly, to address the insufficient real-time performance of data acquisition, a multi-modal sensor array is introduced to achieve high-frequency data capture. Secondly, to address the low efficiency of data processing, an anomaly detection and hierarchical processing mechanism needs to be established to reduce invalid data transmission. Thirdly, to solve the disconnect between control commands and execution, a precise conversion interface from digital signals to physical actions needs to be designed. Fourthly, to address the challenge of heterogeneous equipment access, a multi-protocol compatible actuator interface should be developed to achieve unified control. Finally, a cloud-based collaborative architecture connects the sensing, analysis, decision-making, and execution stages to form a dynamically optimized closed-loop control system.

[0082] The data acquisition module refers to the sensor network deployed at key nodes of industrial equipment, specifically implemented using a multimodal array composed of vibration and current sensors to achieve comprehensive perception of equipment operating status. The data processing module is a cloud computing unit with data cleaning and feature extraction capabilities, specifically implemented using sliding window normalization and Grubbs criterion anomaly detection algorithms to ensure the validity of data input to the decision-making layer. The decision algorithm module is an intelligent decision-making unit with a built-in optimization calculation model, specifically implemented using multi-objective optimization algorithms to balance key parameters such as energy consumption and efficiency. The execution module is a control terminal with signal conversion and interface adaptation functions, specifically implemented using PWM signal conversion circuits and programmable interface controllers to solve the problem of accurate conversion of digital instructions to physical actions.

[0083] The system collects real-time equipment operation data through multimodal sensors, transmits it to the data processing module via the cloud for standardization and outlier removal, and generates high-quality data reports. The decision algorithm module analyzes data characteristics based on an optimization model to generate control commands that balance efficiency and energy consumption. The execution module converts digital commands into PWM control signals via a signal conversion circuit, and then adapts them to different actuators via a programmable interface. The motor interface controls motor speed by adjusting the pulse width, the pneumatic valve interface controls valve opening via level signals, and the hydraulic device interface regulates oil pressure output via current signals. All modules interact and transmit commands through the cloud platform, forming a complete closed loop from status perception to equipment control.

[0084] Compared to existing technologies, traditional systems employ independently operating acquisition and control units, resulting in data flow interruptions. This solution achieves seamless inter-module integration through a cloud-based collaborative architecture, dynamically matching data acquisition frequency with control response speed. While conventional execution units only support a single interface type, this system is equipped with programmable multi-protocol interfaces, adapting to different actuators through signal conversion circuits. Existing data processing workflows lack a hierarchical mechanism; this system establishes pre-processing for anomaly detection and data priority classification, effectively reducing the cloud computing load. Traditional control command generation relies on fixed threshold judgments; this system employs a multi-objective optimization algorithm to achieve dynamic decision-making, improving the matching degree between control strategies and real-time operating conditions.

[0085] The data acquisition module is used to collect raw data on the operating status of industrial equipment and transmit the raw data to the data processing module via the cloud; the data acquisition module collects the current waveform and vibration spectrum of the industrial equipment during operation as raw data.

[0086] The data acquisition module is deployed at key nodes of the industrial equipment and uses a multimodal sensor array to collect equipment operation data. This array includes vibration sensors and current sensors. The vibration and current sensors collect equipment operation data in real time. The vibration sensor, in this embodiment, is set to a high frequency of 10kHz to collect equipment vibration signals, capturing subtle changes in the equipment's mechanical vibration. The current sensor acquires the motor's operating current signal in real time, reflecting electrical parameters such as motor load. The collected raw analog signals are converted into digital signals by the sensor's built-in analog-to-digital converter, serving as raw data. This raw data is then packaged and transmitted to the cloud-based data processing module via an industrial IoT communication protocol.

[0087] Among these, the current waveform refers to the current signal curve that changes over time during motor operation. This can be achieved using a Hall effect current sensor in conjunction with a sampling circuit, capturing the current amplitude, phase, and harmonic components to reflect the motor's load state and electrical anomalies. The vibration spectrum refers to the energy distribution of the equipment's mechanical vibration signal in the frequency domain. This can be achieved using a piezoelectric vibration sensor combined with a fast Fourier transform algorithm, used to identify characteristic frequencies of mechanical faults such as bearing wear and abnormal gear meshing. The multimodal sensor array is a collection of heterogeneous sensing units composed of vibration and current sensors. This can be achieved through a distributed deployment to cover key transmission nodes of the equipment, enabling synchronous acquisition of mechanical and electrical parameters. High-frequency acquisition refers to the vibration sensor operating at a sampling frequency higher than conventional polling mechanisms. This can be achieved by configuring a high-speed analog-to-digital converter (ADC), used to capture minute abnormal fluctuations in transient vibration signals. The sensor's built-in analog-to-digital converter (ADC) unit is a signal conditioning circuit integrated into the sensor body. This can be implemented using a 24-bit high-precision ADC chip, used to convert the original analog signal into a digital signal in real time to reduce transmission noise interference.

[0088] Vibration and current sensors are deployed at critical nodes in industrial equipment. Vibration sensors convert mechanical vibrations into electrical signals using the piezoelectric effect, and experienced experts set the sampling frequency to capture high-frequency vibration details, such as transient impact signals generated by the collision of bearing balls and cages. Current sensors measure motor winding current in real time using electromagnetic induction, simultaneously recording harmonic distortion and transient fluctuations in the current waveform. The analog signals output by both sensors are converted into digital signals by a built-in 24-bit ADC, then packaged into data packets using an Industrial Internet of Things (IIoT) protocol and transmitted to a cloud processing module. This solution constructs a multi-dimensional monitoring data source through the coordinated acquisition of mechanical vibration and electrical parameters. High-frequency vibration sampling can identify early mechanical wear, while current waveform analysis can detect potential faults such as motor winding insulation degradation; the combination of these two technologies forms a complementary monitoring dimension.

[0089] Traditional solutions rely on single-parameter acquisition and fixed-period polling, failing to capture transient changes in equipment operating status. For example, conventional vibration monitoring uses sampling frequencies below 1kHz, potentially missing high-frequency vibration characteristics; current monitoring only records average values, ignoring waveform details. This embodiment achieves synchronous high-frequency acquisition of mechanical and electrical parameters through a multi-modal sensor array. The vibration sampling frequency is increased to 10kHz, enabling complete recording of impact signals caused by bearing failures; the current sensor continuously captures waveform details, identifying current distortions caused by sudden load changes. Existing designs that separate the sensor and data acquisition unit are prone to signal attenuation. This embodiment, however, uses a built-in analog-to-digital converter in the sensor, directly digitizing the signal at the sensing end, avoiding signal-to-noise ratio degradation caused by long-distance transmission.

[0090] The data acquisition module addresses the issues of insufficient real-time data acquisition and limited monitoring dimensions in industrial IoT systems. High-frequency vibration sampling accurately captures transient anomalies in the mechanical structure of equipment, current waveform monitoring reflects real-time changes in motor load and electrical faults, and multi-modal data fusion enhances the comprehensiveness of condition monitoring. The integration of the sensor's built-in analog-to-digital converter with industrial IoT protocols ensures complete transmission of high-frequency data, providing a high-fidelity data foundation for cloud processing.

[0091] The data processing module performs outlier detection and hierarchical processing on the received raw data to generate a data report on the operation of industrial equipment on the production line, and then sends the data report to the decision algorithm module.

[0092] The data processing module standardizes the raw data using the sliding window method to obtain time-series data, and performs outlier detection and removal using the Grubbs criterion. A hierarchical processing mechanism is established to predict the lifespan of industrial equipment based on the data after outlier removal, and hierarchical processing is performed based on the predicted lifespan of industrial equipment.

[0093] The sliding window method involves dynamically dividing the original data sequence into fixed-length windows. This dynamic windowing preserves the temporal correlation of the data and eliminates dimensional differences between different devices. Standardization involves normalizing the data within the window, typically by calculating the mean and standard deviation. For example, subtracting the mean from each window's data and dividing by the standard deviation ensures the processed data follows a standard normal distribution. This method eliminates data fluctuations, providing a unified benchmark for subsequent analysis. The Grubbs criterion is a statistical method for detecting outliers. It involves calculating the absolute value of the residual between a data point and the mean and comparing it to a critical value. This removes data points deviating from the normal distribution, preventing outliers from interfering with the lifespan prediction model. The tiered processing mechanism classifies equipment status levels based on lifespan prediction results. This is achieved by setting multi-level thresholds to trigger differentiated control strategies, such as dividing remaining lifespan into normal, warning and adjustment, and shutdown levels. This method optimizes network bandwidth allocation by dynamically adjusting data processing priorities.

[0094] Existing data processing methods typically employ fixed-time intervals for data cleaning, which fails to adapt to dynamic changes in equipment operating status, resulting in insufficient sensitivity for anomaly detection. Conventional lifetime prediction models are trained directly using raw data, neglecting the impact of data distribution differences on model accuracy. This embodiment eliminates data heterogeneity through sliding window standardization and improves data quality through iterative anomaly detection, ensuring temporal consistency and statistical reliability of the input data for the lifetime prediction model. A tiered processing mechanism dynamically adjusts the data transmission strategy based on the prediction results; for example, under normal conditions, only key feature data is uploaded, reducing network load compared to traditional full data transmission methods.

[0095] The sliding window method determines the size w (the number of data points contained within the window) and the sliding step size (the number of data points the window moves in each step) of the sliding window based on the sampling frequency and data characteristics of the original data. Starting from the beginning of the original data sequence, multiple data windows are sequentially divided according to the set window size and sliding step size, covering the entire original data sequence. For each divided data window, a standardization formula is used to process the data within the window, transforming it into time-series data with uniform distribution characteristics. The processed data is retained in the original window position, forming a standardized time-series data sequence. This method preserves the temporal correlation of data by dynamically dividing windows and eliminates dimensional differences between different devices.

[0096] The standardized formula, the calculation formula is as follows: .

[0097] in, This represents the value of the original data at time t, that is, the original value of the t-th data point within the current window.

[0098] It is the mean of the data within the current sliding window w, calculated using the following formula: i is any natural number. This is the value of the i-th data point within the window, and this parameter reflects the average value of the data within the window; It is the standard deviation of the data within the current sliding window w, calculated using the following formula: It reflects the degree of dispersion of data within the window relative to the mean; the larger the standard deviation, the more drastic the fluctuation of the data.

[0099] After the standardization process using the sliding window method described above, the original data is transformed into time series data with uniform mean and standard deviation, which facilitates subsequent operations such as outlier detection, data analysis, and model training, effectively improving the accuracy and effectiveness of data processing.

[0100] The outlier detection employs the Grubbs criterion. The Grubbs criterion, a branch of the normal distribution, states that if the absolute value of the residual error of a measurement exceeds the critical value of the Grubbs criterion, then that value is considered to have a large error and should be discarded. The Grubbs criterion specifically includes:

[0101] Time series data As sample data, denoted as Where n is the number of sample data, This is the nth sample data.

[0102] For any data Calculate its Grubbs statistic. The calculation formula is: .

[0103] Where i is any natural number, It is the sample mean, calculated using the following formula: s is the sample standard deviation, calculated using the formula: .

[0104] Based on the number of sample data n and the set significance level α, the Grubbs critical value is obtained by looking up a table. .

[0105] Compare the Grubbs statistic for each data point With critical value ,like Then determine This is an outlier.

[0106] Remove all data deemed abnormal in real time.

[0107] Recalculate the mean and standard deviation for the data after removing outliers, and repeat the above steps until all data are processed. The result is considered normal.

[0108] Time-series data, after being standardized using a sliding window, forms a sample dataset. For each data point, its Grubbs statistic is calculated sequentially. The statistic reflects the degree of deviation of the data point from the current data distribution. By comparing this statistic with a critical value obtained through dynamic lookup, a significant anomaly is determined. If the statistic exceeds the critical value, the data point is marked as an outlier and removed. Subsequently, the mean and standard deviation are recalculated based on the remaining data, the data distribution parameters are updated, and outlier detection is performed again, forming a dynamic iterative process. This cyclical calculation can gradually correct the statistical characteristics of the dataset, preventing outliers from interfering with subsequent equipment life prediction models. For example, in motor vibration signal analysis, abnormal vibration peaks may be caused by transient interference. Through multiple iterations, such interfering data can be effectively identified and removed, preserving the signal characteristics that reflect the true state of the equipment.

[0109] Traditional anomaly detection methods typically employ fixed thresholds or simple standard deviation multiples, failing to dynamically adjust detection criteria based on data distribution. In contrast, the Grubbs criterion constructs a dynamic critical value based on statistical theory, combining sample size and significance level for scientific judgment, avoiding the subjectivity of manually set thresholds. Existing technologies do not update statistical parameters after a single detection, leading to residual outliers affecting subsequent analysis. This embodiment continuously optimizes data distribution parameters through iterative calculations, ensuring complete removal of outliers and solving the problem of incomplete detection in traditional methods.

[0110] The industrial equipment life prediction specifically includes:

[0111] The time-series data after outlier removal is converted into a frequency domain signal using a Fast Fourier Transform (FFT) to extract frequency components. Then, a Short-Time Fourier Transform (SFT) is used to capture the changes in frequency components over time, extracting time-varying features and calculating the power spectral density to obtain the time-varying features of prominent power distribution faults. The contribution of each time-varying feature to equipment life prediction is calculated, and the most predictive features are selected as frequency domain features. Based on these frequency domain features, a convolutional neural network is used to build an industrial equipment life prediction model in the cloud. Newly acquired data is processed in real time to predict remaining life.

[0112] Among them, Fast Fourier Transform (FFT) refers to a mathematical transformation that decomposes a time-domain signal into different frequency components. Specifically, it can be implemented using a fast algorithm of Discrete Fourier Transform (DFT) to overcome the limitations of time-domain analysis and reveal the characteristic frequencies of equipment faults. Short-Time Fourier Transform (SFT) is a time-frequency analysis method that performs windowed segmentation of the signal. Specifically, it can use the Hanning window function to extract local signal segments to capture the frequency drift phenomenon of fault characteristics such as bearing wear. Power spectral density refers to the distribution intensity of signal power in the frequency domain, which can be calculated by squared Fourier transform coefficients and used to quantify the energy distribution of fault characteristics. Mutual information is an index that measures the statistical correlation between two variables. Specifically, it can be calculated by the logarithm ratio of the joint probability to the marginal probability and used to screen time-varying features that contribute the most to lifespan prediction. Convolutional Neural Networks (CNNs) are deep learning models with local connectivity and weight sharing characteristics. Specifically, they can use two-dimensional convolutional kernels to process frequency-domain feature tensors and automatically extract fault evolution patterns in the frequency-time dimension.

[0113] After the time-domain vibration signal is converted into a frequency-domain signal using a Fast Fourier Transform (FFT), the characteristic frequency components of the bearing outer ring fault are identified. A Short-Time Fourier Transform (SFT) segments the signal using a sliding time window to generate a time-frequency matrix, capturing the shift in characteristic frequencies caused by increasing wear. Power spectral density calculation quantifies the energy distribution of each frequency component. Combined with mutual information, the correlation between different time-varying characteristics and equipment lifespan is assessed, and spectral centroid drift and instantaneous energy mutations are selected as key frequency-domain features. A convolutional neural network performs sliding operations on the time-frequency matrix using convolution kernels to extract frequency component variation trends and abnormal peak patterns. After dimensionality reduction via pooling layers, these are mapped to predicted remaining lifespan values ​​via fully connected layers. Newly acquired data is input into the trained model after undergoing the same preprocessing steps, enabling real-time output of equipment health status assessment results.

[0114] Traditional methods rely on a single threshold to determine equipment status, only detecting transient anomalies and failing to capture progressive fault characteristics. Conventional frequency domain analysis uses Fourier transforms over fixed time periods, making it difficult to identify frequency drift phenomena in time-varying faults such as bearing wear. Existing life prediction models are mostly based on time-domain statistical features, lacking in-depth mining of time-varying frequency domain characteristics.

[0115] The data processing module achieves deep integration of multi-dimensional operating parameters, accurately captures dynamic feature changes in the equipment degradation process through time-frequency joint analysis, and automatically extracts fault evolution patterns using convolutional neural networks. This effectively overcomes the shortcomings of single threshold alarm mechanisms in recognizing complex fault modes and significantly improves the accuracy of industrial equipment remaining life prediction.

[0116] The Fast Fourier Transform converts time-series data into frequency-domain signals. The Fast Fourier Transform is a mathematical transformation method that decomposes time-domain vibration or current signals into frequency components. Specifically, it can be implemented using the Discrete Fourier Transform algorithm to reveal abnormal characteristic frequency components corresponding to equipment faults.

[0117] After equipment vibration or current signals are converted into frequency domain signals via Fast Fourier Transform (FFT), abnormal amplitudes of characteristic frequency components of mechanical faults such as bearing wear can be highlighted. FFT converts time-series data into frequency domain signals, specifically including:

[0118] Time series data Convert to frequency domain signal The formula is expressed as .

[0119] Where t represents the time point of the sampling point in the time series data, It is a time-domain signal, where Ts is the total number of samples and k is the frequency parameter. The frequency domain amplitude is the signal strength corresponding to frequency component k. The Fast Fourier Transform (FFT) converts time-domain vibration / current and other time-series signals into the frequency domain, revealing the signal's frequency components. Industrial equipment faults often manifest as abnormalities in specific frequency components. This embodiment uses motor bearing wear as an example. When the bearing wears, the vibration signal is at the bearing's characteristic frequency; at the outer ring fault location, the frequency is typically [missing value]. An increase in amplitude may occur, which can be identified using FFT.

[0120] The extraction of time-varying features using the short-time Fourier transform refers to a processing method that performs segmented frequency domain analysis on non-stationary signals through a sliding window. Specifically, it can be implemented using the Hanning window function. The short-time Fourier transform segments the signal through the Hanning window function to generate a time-frequency matrix containing instantaneous energy and spectral centroid, thereby quantifying the dynamic changes in energy distribution during equipment degradation.

[0121] The short-time Fourier transform is used to extract continuous time-varying features. The calculation formula is as follows:

[0122]

[0123] in, It is time-series data; It is a time window function used to extract data from time windows. The local signal segment centered on the Hanning window is used. Tn is the signal length; It is the frequency domain basis function of the Fourier transform, which converts the time-domain signal into a component with frequency k; is the time parameter, i.e., the center position of the window, and k is the frequency parameter.

[0124] Instantaneous energy can be obtained at this time. and spectral centroid As a time-varying feature.

[0125] Instantaneous energy and spectral centroid The calculation formula is: , ;

[0126] Short-Time Fourier Transform (STFT) is a core tool for processing non-stationary signals. It divides the time-domain signal into locally stationary segments by using a sliding window, and then performs frequency domain analysis on each segment to capture the time-varying characteristics of frequency components. By applying a windowing function to the signal segments and performing FFT, a time-frequency matrix is ​​generated. Features that reflect equipment condition degradation are extracted from the STFT time-frequency matrix, capturing the changes in frequency components over time, i.e., the gradual change of characteristic frequencies during fault development. It is suitable for extracting time-varying fault features and highlights the energy of fault characteristic frequencies.

[0127] For each time window, extract the frequency with the largest amplitude. This generates a frequency-time curve, which is used to capture the drift of fault characteristic frequencies, such as the change in characteristic frequency as wear intensifies during bearing failure.

[0128] The power spectral density, i.e., the power distribution at frequency k of the signal in the frequency domain, is calculated using the following formula:

[0129]

[0130] Constructing the time-varying feature matrix

[0131] Where i and j are arbitrary natural numbers, and q is the type of fault. It is the time-varying characteristic matrix of the q-th type of fault in industrial equipment. It's the first frequency magnitude. It is the magnitude of the i-th frequency. The first one The central time window The j-th The central time window

[0132] The mutual information method is used to calculate the contribution of time-varying features. The calculation formula is as follows:

[0133]

[0134] Where y is any natural number, It is the total lifespan of industrial equipment. It is a time-varying characteristic matrix The joint probability that the lifetime Y takes the value f and the lifetime Y takes the value y; yes The probability of taking the value f; It is the probability that the lifespan Y takes the value y.

[0135] Mutual information is a calculation method that measures the statistical correlation between feature variables and equipment lifespan. Specifically, it can be achieved by the ratio of joint probability distribution to marginal probability distribution, and is used to screen the time-frequency features that contribute the most to lifespan prediction.

[0136] Furthermore, based on the joint probability distribution of features and lifetimes, statistical correlation is calculated to automatically select time-frequency feature combinations with strong predictive power, eliminating the subjective bias of manual feature selection.

[0137] Take contribution The time-varying characteristics of the two or more highest faults are used as frequency domain feature B. .

[0138] The industrial equipment life prediction model is constructed using a convolutional neural network, which can be implemented by combining two-dimensional convolutional layers and max pooling layers. This model is used to extract the correlation patterns between frequency and time dimensions from the time-frequency feature matrix. The convolutional neural network applies a two-dimensional convolutional kernel to the time-frequency feature matrix, simultaneously capturing the gradual trends and sudden peaks of frequency components under a sliding window mechanism. After retaining key features through max pooling, the remaining life prediction value is mapped to a fully connected layer, forming an end-to-end equipment health status assessment system.

[0139] Inputting frequency domain features into a convolutional neural network first converts the frequency domain features... As input, frequency domain features are transformed into tensors. Two-dimensional convolutional kernels are used to convolve these tensors to extract local patterns, including trends in fault frequency components and abnormal peaks. A sliding window captures the correlation features between the frequency and time dimensions. The pooling layer uses max pooling to reduce feature dimensionality, decreasing computation while preserving key features. The pooled feature map is flattened into a one-dimensional vector and input to a fully connected layer. This layer integrates the local features extracted by convolution into global features, mapping them to the target space for device lifetime prediction. For Remaining Life (RUL) prediction, the output layer is typically a single-node linear layer, mapping abstract features to specific lifetime prediction values. .

[0140] The tiered processing mechanism is based on the remaining lifespan predicted by the industrial equipment lifespan prediction model. Calculate the health status of industrial equipment The calculation formula is: Y represents the total lifespan of the industrial equipment.

[0141] Set three-level processing thresholds, set thresholds ;

[0142] when Interlock shutdown, when Normal operation at times, when and Maintenance alerts were sent to cloud users.

[0143] when At this time, it is necessary to adjust the equipment load. Normal operation at times, when and Generates running data reports every hour.

[0144] Among these, health level refers to the quantification of equipment status by the ratio of remaining lifespan to total lifespan. Specifically, it can be achieved by using the ratio of the predicted remaining lifespan to the equipment's design lifespan. This parameter dynamically reflects the degree of equipment performance degradation. The three-level processing threshold refers to the maintenance response levels divided according to the equipment's operational safety margin. This can be set by combining historical equipment fault data statistics with expert experience. For example, the first threshold can be set to a health level of 0.2, the second threshold to 0.6, forming three maintenance response intervals. Interlock shutdown refers to triggering the equipment protection program when the equipment health level falls below the minimum threshold. This can be achieved through the safety interlock module in the industrial control system, directly cutting off the equipment's power source and activating the braking device. Adjusting equipment load refers to dynamically adjusting operating parameters within the medium health level range. This can be achieved by using a frequency converter to adjust the motor speed or by using a PLC to control the opening of pneumatic valves, allowing the equipment to maintain basic operation under reduced load. Maintenance early warning generation refers to proactively pushing maintenance requests when the health level reaches a critical point. This can be achieved by sending a structured data packet containing the equipment number, predicted lifespan, and maintenance suggestions to a designated terminal through the cloud platform's message push interface.

[0145] Equipment health status is calculated in real time by the ratio of the remaining lifespan value output by the lifespan prediction model to the equipment's total designed lifespan. When this ratio continuously decreases to the first threshold, it indicates that the equipment is in a high-risk state, and an interlock shutdown is immediately triggered to prevent the fault from escalating. When the health status is in the intermediate range, the system automatically reduces the equipment's operating load, for example, adjusting the motor's rated power to 80%, and records the load adjustment parameters for maintenance reference. When the health status is higher than the second threshold, the system generates a preventative maintenance work order and schedules regular maintenance, such as performing lubrication maintenance after the equipment has run for 500 hours. When the health status approaches the first threshold, the system sends early warning information to maintenance personnel's mobile terminals in advance; for example, a yellow warning is pushed when the health status reaches 0.25, prompting that maintenance should be arranged within 24 hours.

[0146] Traditional solutions using fixed maintenance cycles fail to identify the actual degree of equipment wear and tear. For example, maintenance shutdowns every three months result in over-maintenance of normal equipment while high-risk equipment is not addressed in a timely manner. This embodiment, through a dynamic hierarchical processing mechanism, can match different maintenance strategies based on the real-time health status of the equipment. For example, load adjustment is implemented for equipment with a health level of 0.4 instead of forced shutdown, maintaining production continuity while slowing down equipment degradation. Existing single-threshold alarm mechanisms can only trigger alarms when equipment exceeds limits, while this embodiment achieves a progressive response from early warning to intervention through a three-level threshold division. For example, maintenance resources are prepared as early as a health level of 0.6, significantly shortening fault response time.

[0147] When the decision algorithm module receives a data report and needs to adjust the equipment load, it generates a corresponding control command using a preset algorithm and sends the control command to the execution module. The preset optimization algorithm in the decision algorithm module includes a multi-objective optimization algorithm. The multi-objective optimization algorithm takes energy consumption and efficiency as objectives, establishes a vector space including energy consumption parameters and efficiency parameters, solves the extreme points of the Lagrange function under multi-objective constraints as the optimal solution, and generates control commands based on the optimal solution.

[0148] Multi-objective optimization algorithms are computational methods that handle multiple conflicting optimization objectives through mathematical modeling. Specifically, they can be implemented using non-dominated sorting genetic algorithms or particle swarm optimization algorithms to establish a dynamic balance between minimizing energy consumption and maximizing operational efficiency.

[0149] This embodiment uses energy consumption and efficiency to construct the target model. The vector space of energy consumption parameters and efficiency parameters refers to quantifying the energy consumption index and production efficiency index of equipment operation into a multi-dimensional vector. Specifically, a two-dimensional coordinate system can be constructed using the normalized power consumption value and the output value per unit time to form a multi-objective optimization target space.

[0150] Multi-objective constraints adopt physical limitations that must be met during system operation. Specifically, they can be expressed in the form of linear inequalities, such as the maximum load threshold of computing nodes, the upper limit of network bandwidth, and the task processing time limit, to ensure that the optimization results meet the requirements of actual working conditions.

[0151] Finding the extrema of the Lagrange function involves transforming a constrained optimization problem into an unconstrained one by introducing Lagrange multipliers. Specifically, gradient descent or Newton's iteration method can be used to calculate the saddle point of the objective function in the feasible region, thereby obtaining a solution set that satisfies the Pareto optimality condition.

[0152] When using industrial equipment as nodes and establishing a dynamic task scheduling mechanism between computing nodes and cloud servers, the real-time CPU utilization, memory usage, and network transmission latency of each node are first quantified as efficiency parameters, while the power consumption and heat dissipation of the equipment are quantified as energy consumption parameters. By constructing a vector space containing these two dimensions, the task scheduling problem is transformed into a mathematical problem of finding the optimal point for energy efficiency. In the multi-objective constraint setting stage, the maximum computing power of the nodes, task deadlines, and data transmission bandwidth limits are transformed into a system of linear inequality constraint equations. The Lagrange multiplier method is used to construct a combined function of the objective function and the constraints, and the extreme point of this function in the feasible region is calculated by differentiation. The task allocation scheme corresponding to this extreme point can ensure that the task processing efficiency meets the real-time requirements while keeping the overall energy consumption at a minimum. The finally generated dynamic control instructions will automatically adjust the task distribution strategy according to real-time load fluctuations, for example, prioritizing the allocation of computationally intensive tasks to idle edge nodes during periods of high load on the cloud server.

[0153] Traditional static task allocation strategies use fixed thresholds to divide computational tasks, which cannot respond to real-time load fluctuations on servers and edge nodes. This embodiment solves the inherent contradiction between resource utilization and response latency by constructing a dynamic optimization model that recalculates the optimal allocation scheme for each task scheduling. Compared to single-objective optimization methods, the introduction of a multi-objective vector space can simultaneously optimize energy consumption and efficiency indicators, avoiding system imbalance caused by unilaterally pursuing a single indicator.

[0154] The execution module receives a control command and converts it into a physical control signal that can be executed by the industrial equipment. The execution module converts the PWM control signal of the control command into a physical control signal to adjust the operating state of the industrial equipment.

[0155] The actuator is equipped with various types of actuator interfaces, including motor interfaces, pneumatic valve interfaces, and hydraulic device interfaces, supporting the connection of various physical control units such as motors, pneumatic valves, and hydraulic devices.

[0156] The displacement sensor measures the mechanical displacement of the actuator, typically using a linear variable differential transformer or photoelectric encoder. It converts physical displacement into an electrical signal, directly reflecting the deviation between the actual displacement of the actuator and the control command. The torque feedback signal measures the torque transmitted from the motor output shaft, typically using a strain gauge torque sensor or magnetoelastic torque sensor. It quantifies the actual force applied at the load end to identify mechanical transmission errors and external disturbances. The time-series dataset is a collection of displacement and torque signals arranged chronologically. It integrates multi-source signals into structured data using timestamp alignment, recording dynamic changes during execution to provide a basis for time-series correlation analysis. Closed-loop control is a control mechanism that feeds back the execution results to the data processing module. It compares the monitoring results with the original control commands via a data bus, generates deviation correction parameters, and iteratively optimizes the decision-making algorithm.

[0157] The monitoring module collects the actual displacement of the actuator through displacement sensors and simultaneously acquires the motor output torque signal. These two types of signals are aligned by timestamps to form a time-series dataset. This dataset contains the dynamic trajectory of displacement and torque changes during execution. By comparing the difference between the expected displacement curve and the actual displacement curve, the execution deviation is quantified. The data processing module performs sliding window standardization on the deviation data, extracting characteristic parameters such as deviation amplitude, fluctuation frequency, and cumulative error, which are then input into the decision algorithm module to generate compensation commands. The compensation commands, through the execution module, adjust the duty cycle or amplitude of the PWM signal to correct the actuator output displacement, forming a closed-loop control loop from command generation to execution verification.

[0158] In this embodiment, the displacement sensor can be a laser displacement sensor with a resolution down to the micrometer level, for example, configured at the end of the piston rod of a hydraulic cylinder to measure the actuator displacement in real time; the torque feedback signal can be acquired through a flange-type torque sensor, for example, installed at the connection between the motor output shaft and the reducer, to monitor load torque fluctuations. The time series dataset can use a 5-millisecond sampling interval, for example, storing the displacement signal and torque signal as a two-dimensional array according to timestamps, and transmitting it to the cloud analysis platform through data packet encapsulation.

[0159] Traditional industrial control systems typically only monitor whether the actuator receives commands, lacking real-time feedback on actual displacement and load torque. This results in the inability to promptly correct execution deviations caused by mechanical transmission errors or sudden load changes. This embodiment constructs a time-series dataset by synchronously acquiring displacement and torque signals, enabling the identification of dynamic anomalies such as execution delays, oscillations, and overloads. Combined with a closed-loop control mechanism, it achieves deviation compensation, solving the problem of unverifiable execution effects in open-loop control.

[0160] The execution module enables real-time quantitative evaluation of the control effect, accurately identifies execution deviations caused by factors such as mechanical transmission backlash and load disturbances, and drives dynamic adjustment of control parameters through data feedback. This mechanism effectively improves system control accuracy, avoids equipment malfunctions caused by the accumulation of execution deviations, and provides a data foundation of actuator health status for predictive maintenance.

[0161] The monitoring module monitors the execution of control commands by the execution module in real time as the monitoring result, tracks command execution deviations, and feeds the monitoring results back to the data processing module for subsequent data analysis to form a closed-loop control. The execution status is obtained by collecting displacement signals from displacement sensors and torque feedback signals from motor outputs generated by the execution module in accordance with the control commands in real time, and constructing a time series dataset of the control command execution status as the monitoring result.

[0162] In Example 2:

[0163] The second embodiment of the present invention differs from the first embodiment in that:

[0164] When the aforementioned functions are implemented as software functional units and can be sold or used as independent products, they can be stored in a storage medium that can be read by a computer. Based on this, the technical solution of the present invention, essentially either its contribution to the prior art or a portion of the solution, can be presented in the form of a computer software product.

[0165] The software product is stored in a storage medium and contains a series of instructions that enable computer devices, such as personal computers, servers, or network devices, to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium encompasses various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, and optical disks.

[0166] The logic and steps presented in a flowchart, or otherwise described, can be viewed as an ordered set of executable instructions that perform logical functions. They can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a processor-containing system, or other system that can retrieve and execute instructions from, or in conjunction with, such system, apparatus, or device). Within the scope of this specification, "computer-readable medium" means any means capable of containing, storing, communicating, propagating, or transmitting programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0167] The following are more specific examples of computer-readable media, but this is not exhaustive: electrical connections with one or more wires (belonging to electronic devices), 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 capable of printing programs. This is because programs can be acquired electronically by optically scanning paper or other media, then editing, interpreting, or otherwise appropriately processing them, and subsequently stored in computer memory.

[0168] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be achieved by software or firmware stored in memory and executed by an adapted instruction execution system. For example, if hardware implementation is adopted, as in another embodiment, it can be accomplished using any one or a combination of the following technologies well known in the art: discrete logic circuits with logic gates that implement data signal logic functions, application-specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0169] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An information sensing and control system based on industrial Internet of Things (IoT) cloud services, characterized in that, include: The system comprises a data acquisition module, a data processing module, a decision-making algorithm module, and an execution module. in, The data acquisition module is used to collect raw data on the operating status of industrial equipment and transmit the raw data to the data processing module via the cloud. The data processing module performs outlier detection and hierarchical processing on the received raw data to generate a data report on the operation of industrial equipment on the production line, and sends the data report to the decision algorithm module. The decision algorithm module generates corresponding control commands based on the received data reports using a preset algorithm, and sends the control commands to the execution module. The execution module receives control commands and converts them into physical control signals that can be executed by the industrial equipment to adjust the operating status of the industrial equipment. The data processing module uses the sliding window method to standardize the raw data to obtain time series data, and uses the Grubbs criterion to detect and remove outliers. It also establishes a hierarchical processing mechanism to predict the lifespan of industrial equipment based on the data after outlier removal and performs hierarchical processing based on the predicted lifespan of industrial equipment. The industrial equipment life prediction specifically includes: The time series data after outlier removal is converted into a frequency domain signal using a fast Fourier transform to extract frequency components. Then, the frequency components of the frequency domain signal are captured by short-time Fourier transform to extract time-varying features and calculate power spectral density to obtain the time-varying features of power distribution protrusion faults. Calculate the contribution of each time-varying feature to the equipment life prediction, and select the feature with the most predictive power as the frequency domain feature; Based on frequency domain characteristics, a convolutional neural network is used to build an industrial equipment life prediction model in the cloud. Process newly acquired data in real time to predict remaining lifespan; The Fast Fourier Transform converts time-series data into frequency domain signals, specifically including: converting time-series data... Convert to frequency domain signal The formula is expressed as ; Where t represents the time point of the sampling point in the time series data, It is a time-domain signal, where Ts is the total number of samples and k is the frequency parameter. The frequency domain amplitude is the signal strength corresponding to the frequency component k. The extraction of time-varying features specifically includes: The short-time Fourier transform is used to extract continuous time-varying features. The calculation formula is as follows: in, It is time-series data; It is a time window function used to extract data from time windows. The local signal segment centered on the Hanning window is used. Tn is the signal length; It is the frequency domain basis function of the Fourier transform, which converts the time-domain signal into a component with frequency k; is the time parameter, i.e., the center position of the window; k is the frequency parameter. Instantaneous energy can be obtained at this time. and spectral centroid As a time-varying characteristic; Instantaneous energy and spectral centroid The calculation formula is: , ; The power spectral density, i.e., the power distribution at frequency k of the signal in the frequency domain, is calculated using the following formula: Constructing the time-varying feature matrix Where i and j are arbitrary natural numbers, and q is the type of fault. It is the time-varying characteristic matrix of the q-th type of fault in industrial equipment. It's the first frequency magnitude. It is the magnitude of the i-th frequency. The first one The central time window The j-th The central time window The mutual information method is used to calculate the contribution of time-varying features. The calculation formula is as follows: Where y is any natural number, It is the total lifespan of industrial equipment. It is a time-varying characteristic matrix The joint probability that the lifetime Y takes the value f and the lifetime Y takes the value y; yes The probability of taking the value f; It is the probability that the lifespan Y takes the value y; Take contribution The time-varying characteristics of the two or more highest faults are used as frequency domain feature B. ; The tiered processing mechanism is based on the remaining lifespan predicted by the industrial equipment lifespan prediction model. Calculate the health status of industrial equipment The calculation formula is: Y represents the total lifespan of the industrial equipment. Set three-level processing thresholds, set thresholds ; when Interlock shutdown, when Normal operation at times, when and Maintenance alerts were sent to cloud users. when At this time, it is necessary to adjust the equipment load. Normal operation at times, when and Generates running data reports every hour.

2. The information sensing and control system based on industrial IoT cloud services according to claim 1, characterized in that, The data acquisition module collects the current waveform and vibration spectrum of the industrial equipment as raw data during operation; The data acquisition module uses a multimodal sensor array to collect equipment operation data, and the multimodal sensor array includes vibration sensors and current sensors.

3. The information sensing and control system based on industrial IoT cloud services according to claim 1, characterized in that, The outlier detection employs the Grubbs criterion, specifically including: Time series data As sample data, denoted as Where n is the number of sample data, This is the nth sample data; For any data Calculate its Grubbs statistic. The calculation formula is: ; Where i is any natural number, It is the sample mean, calculated using the following formula: s is the sample standard deviation, calculated using the formula: ; Based on the number of sample data n and the set significance level α, the Grubbs critical value is obtained by looking up a table. ; Compare the Grubbs statistic for each data point With critical value ,like Then determine This is an outlier; Remove all data that is deemed abnormal; Recalculate the mean and standard deviation for the data after removing outliers, and repeat the above steps until all data are processed. .

4. The information sensing and control system based on industrial IoT cloud services according to claim 1, characterized in that, The decision algorithm module includes a multi-objective optimization algorithm. The multi-objective optimization algorithm establishes a vector space including energy consumption parameters and efficiency parameters, solves for the extreme points of the Lagrange function under multi-objective constraints as the optimal solution, and generates control commands based on the optimal solution.

5. The information sensing and control system based on industrial IoT cloud services according to claim 1, characterized in that, It also includes a monitoring module, which monitors the execution of control commands by the execution module in real time as the monitoring result, tracks command execution deviations, and feeds the monitoring result back to the data processing module; The execution status is obtained by collecting displacement signals from displacement sensors and torque feedback signals from motor outputs generated by the execution module in accordance with control commands in real time, and constructing a time series dataset of control command execution status as the monitoring result.

6. An information sensing and control method based on industrial IoT cloud services, characterized in that, The method, applied to the information sensing and control system based on industrial IoT cloud services as described in any one of claims 1 to 5, includes the following steps: Step S1: Collect raw data on the operation of industrial equipment through the data acquisition module; Step S2: Transmit the raw data to the data processing module for data cleaning, processing and analysis to generate a data report; Step S3: Input the data report into the decision algorithm module and generate control commands using a preset algorithm; Step S4: The execution module receives the control command, executes the command according to the control command, and regulates the operating status of the corresponding industrial equipment.

Images