A filter cloth working condition data acquisition and early warning system based on internet of things and edge computing

Abstract

The application provides a filter cloth working condition data acquisition and early warning system based on Internet of Things and edge computing, relates to the technical field of filter cloth working condition monitoring, and comprises a perception layer, an edge computing layer and a cloud management layer.The perception layer comprises a multidimensional sensor array for collecting filter cloth differential pressure, temperature, instantaneous flow, vibration frequency and surface image multi-modal working condition data.The edge computing layer is deployed in an industrial field and integrates a front-end processing module, a lightweight inference engine, a soot removal strategy optimization engine, a lightweight digital twin mirror image and a dual-mode adaptive networking module.The application adopts an edge front-end processing double-layer architecture, breaks through the link limitation of the traditional pure cloud processing mode, all data processing, working condition judgment and early warning output are completed at the edge node in the industrial field, does not need to rely on the cloud server throughout the whole process, compresses the traditional cloud second-level response delay to millisecond level, and simultaneously has complete off-line working capacity in a broken network.

Description

Technical Field

[0001] This invention relates to the field of filter cloth condition monitoring technology, and in particular to a filter cloth condition data acquisition and early warning system based on the Internet of Things and edge computing. Background Technology

[0002] Existing industrial filter cloth condition monitoring systems generally adopt a pure cloud-based centralized processing architecture. They rely on IoT sensors to collect basic data such as pressure difference, temperature, and flow rate of the filter cloth. All raw data is uploaded to the cloud server in real time via the network. The cloud performs data cleaning, feature analysis, condition determination, fault warning, and equipment control command issuance. At the same time, a fixed-cycle timed dust removal strategy is used to achieve dust removal and maintenance of the filter cloth. At the communication level, data transmission is mostly achieved using a single WiFi or cellular network.

[0003] However, the above technologies have the following drawbacks: First, data processing relies entirely on cloud transmission, resulting in long network transmission links and massive data volumes, leading to second-level delays in condition analysis and early warning response. Furthermore, the system directly fails during network outages or fluctuations, making continuous on-site monitoring impossible. Second, cloud-based batch data processing suffers from redundant computing power and a high proportion of invalid data transmission, consuming enormous network bandwidth and cloud storage resources, hindering localized real-time intelligent reasoning. Third, the dust removal strategy uses a uniform timing mode, which cannot be precisely adjusted according to the actual deterioration type of the filter cloth, easily causing excessive dust removal fatigue damage or incomplete dust removal clogging. Fourth, it can only achieve passive alarms after a fault occurs, lacking the ability to predict condition trends, resulting in severe early warning lag. Fifth, the single communication mode has weak resistance to industrial electromagnetic interference, a high data packet loss rate, and continuous high-frequency data acquisition and transmission leads to excessive overall power consumption, resulting in poor adaptability to complex industrial scenarios. Therefore, this invention proposes a filter cloth condition data acquisition and early warning system based on the Internet of Things and edge computing to solve the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a filter cloth condition data acquisition and early warning system based on the Internet of Things (IoT) and edge computing. This system adopts a two-layer architecture with edge-end front-end processing, breaking through the link limitations of traditional pure cloud processing modes. All data processing, condition judgment, and early warning output are completed at the edge nodes in the industrial field, eliminating the need to rely on cloud servers throughout the process. This reduces the traditional cloud-based second-level response latency to millisecond-level, while also possessing complete offline working capabilities. In the event of network anomalies, it can independently complete the entire process of monitoring and early warning, and automatically re-upload synchronized data after reconnecting to the network. This solves the problems of high network dependence, excessive latency, and failure due to network outages in traditional systems, comprehensively ensuring the continuity and stability of industrial filter cloth monitoring.

[0005] To achieve the objectives of this invention, the following technical solution is provided: a filter cloth operating condition data acquisition and early warning system based on the Internet of Things and edge computing, comprising a perception layer, an edge computing layer, and a cloud management layer. The perception layer includes a multi-dimensional sensor array for acquiring multi-modal operating condition data of filter cloth, such as differential pressure, temperature, instantaneous flow rate, vibration frequency, and surface image. The edge computing layer is deployed in the industrial field and integrates a pre-processing module, a lightweight inference engine, a dust removal strategy optimization engine, a lightweight digital twin mirror, and a dual-mode adaptive networking module. It is used to locally complete data cleaning, time-series feature extraction, intelligent operating condition judgment, precise fault control, operating condition trend prediction, and adaptive data transmission. It supports independent operation offline and automatic data synchronization when connected to the network, thus forming an edge-front dual-layer processing architecture. The first layer completes the pre-processing and cleaning of raw data, and the second layer completes intelligent decision-making and command output.

[0006] The cloud-based management layer is used for model iteration, global data storage, and remote device control, enabling millisecond-level real-time monitoring of filter cloth operating conditions, proactive early warning, and precise operation and maintenance control.

[0007] A further improvement lies in the fact that the lightweight inference engine incorporates a cloud-pruned and quantized one-dimensional deep separable convolutional neural network model, which concatenates multimodal operating condition data into a comprehensive feature vector.

[0008] ,

[0009] Where: Xt is the comprehensive feature vector of the filter cloth working condition at time t. denoted as the real-time pressure difference parameter between the front and rear ends of the filter cloth, T as the real-time operating temperature of the filter cloth, Q as the instantaneous flow rate of the filter cloth duct, Vf as the vibration amplitude-frequency feature vector of the filter cloth, and Gt as the surface texture feature vector of the filter cloth at time t.

[0010] A further improvement lies in the fact that the lightweight inference engine extracts temporal features through one-dimensional depthwise separable convolution operations, the convolution operation formula being:

[0011] ,

[0012] Where: Yt is the feature extraction output vector at time t; is a nonlinear activation function; Wd is a depthwise separable convolution weight matrix; * is the convolution operation; b is the convolution bias term, which captures the temporal evolution of filter cloth clogging, scaling, and damage.

[0013] Further improvements are made in that: the lightweight inference engine outputs six types of operational condition confidence scores through a fully connected classification layer, determines anomalies based on preset local alarm thresholds, and when the abnormal operational condition confidence score is... The system triggers an early warning, where Si is the real-time confidence score of the i-th type of operating condition, S0 is the preset fixed alarm threshold of the edge node, and i takes values ​​from 1 to 6, corresponding to six types of operating conditions: normal operation, minor blockage, moderate blockage, severe blockage, filter cloth damage, and surface scaling.

[0014] A further improvement lies in the fact that the lightweight digital twin mirror image uses a particle filter algorithm to predict operating conditions, and the state update formula is as follows:

[0015] ,

[0016] in: This is the predicted value of the filter cloth's operating condition at the next moment. The nonlinear evolution transfer function of the filter cloth working condition is given by St, where St is the real-time three-dimensional working condition value of the filter cloth pressure difference, flow velocity and temperature at time t, and wt is the Gaussian noise parameter of the system process at time t. It is used to iteratively correct the state estimation error and accurately predict the evolution trend of filter cloth resistance in the next 30 minutes.

[0017] Further improvements are made in that the dust removal strategy optimization engine adopts a rule-model hybrid decision-making mechanism to construct a degradation matching formula.

[0018] ,

[0019] Where: P is the optimal dust removal control parameter set, F is the hybrid decision matching function, Ty is the fault type label, and Kx is the preset dust removal rule parameter library, so as to realize differentiated and precise control of blockage, damage, and dust removal failure faults.

[0020] Further improvements are made in that the dual-mode adaptive networking module adopts LoRa+WiFi dual communication modes, collects channel interference intensity, network signal-to-noise ratio and packet loss rate parameters in real time, switches to LoRa transmission mode when the interference intensity is higher than a preset threshold, and switches to WiFi high-speed transmission mode when the network signal is good, so as to realize adaptive anti-interference transmission in complex industrial scenarios.

[0021] Further improvements are made in that: the pre-processing module is a low-level data processing unit, used to receive multimodal raw operating condition data transmitted from the perception layer, and uniformly complete data denoising and filtering, outlier removal, time-series alignment, multi-dimensional data normalization and image data preprocessing operations. It also normalizes and fuses sensor data of different sampling frequencies and dimensions, removes invalid noise data generated by industrial site interference, and outputs standardized and structured time-series operating condition data, providing a high-quality input data source for the lightweight inference engine and lightweight digital twin mirror.

[0022] Further improvements include: the edge computing layer supports an offline working mechanism when the network is disconnected. When the network is disconnected, it can independently complete data collection, analysis, early warning and local data caching. After the network is restored, it automatically verifies missing data and synchronizes it to the cloud in batches to achieve closed-loop archiving of network data. The edge computing layer is equipped with a dynamic sleep-wake mechanism to identify the device's idle and running states in real time. In the idle state, it automatically reduces the data collection frequency and shuts down redundant transmission modules. In the running state, it restores high-frequency collection and high-speed transmission, adaptively reducing the device's operating power consumption.

[0023] Further improvements include: the cloud management layer has global data control and model iteration optimization functions, specifically including lightweight model offline training, pruning quantization iteration optimization, remote configuration and distribution of edge node operating parameters, full-scale operating data archiving and storage, historical fault big data source tracing and analysis, and equipment operation and maintenance statistical report generation functions. At the same time, it receives abnormal data and offline cached data synchronized from edge nodes, completes global data closed-loop aggregation, and continuously optimizes the parameter library of inference and dust removal strategies of the edge computing layer based on massive cloud data, so as to realize the system's adaptive iterative upgrade.

[0024] The beneficial effects of this invention are as follows:

[0025] 1. This invention adopts a two-layer architecture of edge-end front-end processing, breaking through the link limitations of the traditional pure cloud processing mode. All data processing, working condition judgment, and early warning output are completed at the edge nodes in the industrial field, without relying on cloud servers throughout the process. It compresses the traditional cloud-based second-level response latency to the millisecond level, and at the same time has complete offline working capability when the network is abnormal. It can independently complete the entire process monitoring and early warning when the network is abnormal, and automatically re-transmit synchronized data after reconnecting to the network. It solves the problems of high network dependence, high latency, and failure when the network is out of service in traditional systems, and fully ensures the continuity and stability of industrial filter cloth monitoring.

[0026] 2. This invention is equipped with a lightweight inference engine, which optimizes the deep neural network model through pruning and quantization techniques. It completes multimodal feature fusion and intelligent inference of working conditions locally. Data upload is triggered only when abnormal working conditions are detected. Normal working condition data is processed locally without uploading, which greatly reduces the amount of invalid data transmission, reduces the bandwidth occupation of IoT transmission and the consumption of cloud storage and computing resources, and avoids the latency loss caused by round-trip data transmission in the cloud. It achieves millisecond-level accurate identification of six types of filter cloth working conditions, and improves the real-time performance and resource utilization of working condition monitoring.

[0027] 3. This invention sets up a hybrid decision-making cleaning optimization engine driven by filter cloth deterioration type, abandoning the traditional uniform timed cleaning mode. It matches exclusive control strategies for three core deterioration faults: clogging, damage, and cleaning failure, to achieve precise intervention and control for each fault. This effectively avoids filter cloth mechanical fatigue and fiber peeling damage caused by excessive cleaning, as well as filter cloth clogging caused by insufficient cleaning. Actual tests have shown that it can effectively extend the service life of filter cloth and reduce equipment operation and maintenance costs and filter cloth replacement frequency.

[0028] 4. This invention constructs a lightweight digital twin image of filter cloth, retaining only the core state variables to complete the real-time mapping of the three-dimensional state space. Relying on the particle filter algorithm, it accurately predicts the evolution trend of filter cloth resistance and generates a fault warning window in advance. This completely changes the passive alarm mode after the traditional system fails, realizing a technological leap from passive handling to active prediction. The early warning advance is more than 3 times that of the traditional system, leaving sufficient time for on-site operation and maintenance and equipment control.

[0029] 5. This invention designs a LoRa+WiFi dual-mode adaptive IoT networking mechanism, which can sense channel interference and network signal quality in the industrial field in real time, automatically switch to the optimal communication mode, and take into account both high-speed transmission and anti-interference requirements. At the same time, it is equipped with a dynamic sleep-wake mechanism to adaptively adjust the data acquisition and transmission frequency, reduce the power consumption of the device in the idle state, and effectively solve the problems of weak anti-interference, high packet loss rate and high power consumption of traditional single-mode networking. It is suitable for complex industrial electromagnetic interference and harsh scenarios with multiple obstructions, and ensures stable data transmission at all times. Attached Figure Description

[0030] Figure 1 This is a diagram illustrating the composition of the present invention. Detailed Implementation

[0031] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0032] Example 1

[0033] according to Figure 1 As shown, this embodiment proposes a filter cloth operating condition data acquisition and early warning system based on the Internet of Things and edge computing. It is applied to the harsh operating conditions of high temperature, high dust, and frequent network fluctuations in the dust removal workshop of the steel industry. To address the problems of high latency and network failure in traditional cloud monitoring, the present invention deploys an edge dual-layer front-end processing architecture, as detailed below:

[0034] On-site hardware deployment: Edge computing gateway nodes are deployed in each independent dust removal box unit in the workshop. Each edge node corresponds to a set of filter cloth sensor arrays. The sensor arrays include differential pressure sensors, temperature sensors, flow sensors, vibration sensors and high-definition image acquisition modules. All sensors are connected to the edge nodes nearby to realize on-site data collection and access nearby, eliminating the signal attenuation problem of long-distance wiring.

[0035] Dual-layer architecture and layered working mechanism: The system strictly adopts a dual-layer processing architecture at the edge. The first layer is the data preprocessing layer, which continuously receives raw data collected from various sensors and performs parallel data denoising and filtering, outlier removal, multi-source data time-series alignment, and normalization and fusion processing of data with different sampling frequencies. The second layer is the intelligent decision output layer, which receives the preprocessed standardized data and completes the entire process of operating condition feature analysis, status determination, fault warning, and control command generation. All calculation and judgment tasks are completed locally at the edge without cloud involvement.

[0036] Online / offline dual-mode switching mechanism: Under normal online status, edge nodes only synchronize abnormal event logs, critical operating condition extreme data and operation and maintenance records to the cloud, while massive amounts of regular time-series data are stored and processed locally to reduce network pressure; real-time monitoring of network latency, packet loss rate and connection status, and automatic seamless switching to offline independent working mode when abnormalities such as network disconnection, signal fluctuation, or link interruption are detected.

[0037] Offline caching and data completion mechanism: In the offline state, the edge node can independently and continuously complete data collection, analysis, early warning and command control for 72 hours. At the same time, it caches all time-series operating data, equipment operation data and fault record data locally in a large capacity. After the network link is restored to normal, it automatically verifies the missing historical data in the cloud and completes the offline data synchronization and archiving in batches, time-sharing and in an orderly manner to ensure that the entire network data is closed-loop, without loss or error.

[0038] Actual test results: After the implementation of this embodiment, the overall system response delay was stably controlled at 10-20ms, which is two orders of magnitude better than the 1-3s response delay of traditional cloud systems. During the network outage, the equipment monitoring, early warning, and dust removal control functions operated normally throughout, without any monitoring interruption, missed reports, or false reports, thus completely solving the problem of monitoring failure caused by network instability in industrial sites.

[0039] Example 2

[0040] according to Figure 1 As shown, this embodiment proposes a filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing. Relying on a lightweight neural network model deployed at the edge, it achieves millisecond-level accurate identification of six types of filter cloth working conditions, replacing the traditional cloud computing power inference mode, as detailed below:

[0041] Lightweight model deployment process: In the early stage, the original deep neural network is trained on a cloud server based on a massive industrial filter cloth working condition sample dataset. Redundant weights and neurons are removed by pruning the model. The 32-bit floating-point parameters are compressed into 8-bit integer parameters by quantization technology. While ensuring no loss of recognition accuracy, the model size and computing power consumption are greatly reduced. The optimized lightweight one-dimensional deep separable convolutional neural network model is then solidified and deployed to edge computing nodes to adapt to low-computing-power edge hardware environments.

[0042] Multimodal feature fusion process: Edge nodes synchronously collect the pressure difference ΔP between the front and rear ends of the filter cloth, working temperature T, instantaneous flow rate Q of the air duct, filter cloth vibration amplitude-frequency feature vector Vf, and filter cloth surface image texture feature vector Gt in real time. The standardized comprehensive feature vector at time t is generated by splicing according to a fixed dimension, realizing the unified fusion of multi-source heterogeneous data and solving the problem of low accuracy of single feature recognition under working conditions.

[0043] Local temporal feature inference mechanism: One-dimensional depthwise separable convolution operation is used to extract local temporal features from the comprehensive feature vector. It focuses on capturing subtle temporal patterns such as slow changes in pressure gradient, equipment operation vibration spectrum drift, filter cloth surface texture wear, and scale evolution during the filter cloth clogging process. Feature mapping is completed through a fully connected classification layer, and independent confidence scores are output for six operating conditions: normal operation, slight clogging, moderate clogging, severe clogging, filter cloth damage, and surface scale.

[0044] Local threshold warning and data upload strategy: Edge nodes are preset with a fixed alarm confidence threshold of 0.8. When the real-time confidence score of any abnormal working condition exceeds the threshold, the lightweight inference engine immediately generates a local abnormal event signal, triggering audible and visual warnings and equipment control commands. Event data, feature data and on-site image data are uploaded only when an abnormality is determined. Normal working condition data is all processed and stored locally without cloud interaction.

[0045] Actual test results: In this embodiment, the time for single-frame full-dimensional working condition inference is ≤15ms, and the comprehensive accuracy of the six-class working condition recognition reaches 98.7%. Compared with the traditional cloud inference architecture, the amount of invalid data uploaded is reduced by more than 75%, which greatly reduces the network bandwidth occupation and cloud storage and computing power resource consumption, and truly realizes real-time intelligent perception at the edge.

[0046] Example 3

[0047] according to Figure 1 As shown, this embodiment proposes a filter cloth operating condition data acquisition and early warning system based on the Internet of Things and edge computing. Based on a rule-model hybrid decision-making module, it abandons the traditional unified fixed-time dust cleaning mode and realizes differentiated and precise operation and maintenance control for each specific problem, as detailed below:

[0048] Hybrid decision architecture: A dual-drive decision engine of model recognition and rule matching is integrated at the edge computing layer. The model side is responsible for accurately identifying the specific deterioration and fault types of the filter cloth, while the rule side has a built-in preset dust removal parameter library and fault handling logic threshold. It can match the optimal control scheme in real time without manual intervention and achieve fully automatic adaptive control.

[0049] Precise dust removal control for clogging faults: For the three types of faults identified by the model—minor clogging, moderate clogging, and severe clogging—the system adaptively adjusts the dust removal parameters in a gradient manner: for minor clogging, the pulse jet width is extended by 20% and the jet pressure is increased by 0.1MPa; for moderate clogging, the pulse jet width is extended by 35% and the jet pressure is increased by 0.2MPa; for severe clogging, the pulse jet width is extended by 50% and the jet pressure is increased by 0.3MPa, thereby enhancing the dust removal effect and quickly eliminating filter cloth clogging resistance.

[0050] Damage-related fault handling mechanism: When the model identifies fault types such as partial damage or tearing of the filter cloth, the system will no longer perform dust cleaning operation, immediately trigger the local dust removal unit isolation protection mechanism, automatically close the corresponding air inlet valve, stop the filtration work of the unit, and trigger on-site audible and visual alarms and push fault work orders to the cloud to prompt maintenance personnel to shut down the machine for inspection and repair, so as to avoid dust leakage and secondary damage to the equipment due to the continued operation of the damaged filter cloth.

[0051] Self-diagnosis and repair of dust removal failure: For dust removal failures where the working conditions do not improve after dust removal and the differential pressure remains high, the system automatically starts the pulse valve self-diagnosis process, which monitors the pulse valve opening response delay, the stability of the blowing air pressure, and the pipeline air pressure loss parameters in real time. It accurately locates fault points such as valve jamming, pipeline leakage, and insufficient air pressure, and automatically generates diagnostic reports and maintenance suggestions.

[0052] Actual test results: This embodiment completely eliminates the problems of ineffective and excessive blowing caused by traditional timed dust cleaning. The frequency of mechanical fatigue and fiber peeling damage of the filter cloth is reduced by 60%, the average service life of the filter cloth is increased by 42%, and the filter cloth replacement cost and equipment downtime maintenance time are significantly reduced.

[0053] Example 4

[0054] according to Figure 1 As shown, this embodiment proposes a filter cloth condition data acquisition and early warning system based on the Internet of Things and edge computing. By building a miniature digital twin mirror and particle filter prediction algorithm at the edge, it realizes a paradigm upgrade of filter cloth condition from passive alarm to active prediction, as detailed below:

[0055] Lightweight twin modeling solution: Abandoning the traditional industrial digital twin modeling method with full parameters and high computing power, this solution focuses on the core operating characteristics of the filter cloth and retains only three core state variables: pressure difference, airflow velocity, and operating temperature. It builds a lightweight three-dimensional state space mapping model to replicate the filter cloth resistance state and operating conditions in real time. The model is small in size, consumes low computing power, and is fully compatible with the hardware performance of edge nodes.

[0056] Particle filter iterative prediction mechanism: The system constructs an iterative update logic for the working condition based on the particle filter algorithm. Relying on the real-time three-dimensional state data of the filter cloth and historical time series operation data, it continuously iterates and corrects the state estimation error, dynamically fits the evolution law of dust adhesion and resistance growth of the filter cloth, and accurately predicts the trend of filter cloth resistance change and pressure difference over-limit risk in the next 30 minutes.

[0057] Tiered proactive early warning mechanism: Based on the predicted evolution results, two levels of early warning windows are set: Level 1 predictive early warning (20-30 minutes in advance) prompts maintenance personnel to prepare for inspection in advance; Level 2 imminent early warning (5-20 minutes in advance) automatically triggers preventive dust cleaning control commands to intervene in the potential for filter cloth blockage in advance and avoid the actual occurrence of the fault.

[0058] Real-time status update calibration: The digital twin mirror synchronizes the preprocessed standard working condition data at the edge in real time, and completes a status update and algorithm iteration every 100ms. Combined with real-time working conditions, the prediction deviation is corrected, eliminating the problems of lag and insufficient accuracy of static model prediction, and ensuring that the prediction results are consistent with the actual working conditions on site.

[0059] Actual test results: The average early warning lead time of this embodiment can reach 25-30 minutes, and the early warning prediction capability is 3.2 times that of traditional passive alarm systems. It effectively avoids faults such as excessive filter cloth pressure difference, severe blockage, and equipment overload operation, and greatly improves the operational stability and maintenance initiative of dust removal equipment.

[0060] Example 5

[0061] according to Figure 1 As shown, this embodiment proposes a filter cloth operating condition data acquisition and early warning system based on the Internet of Things and edge computing. It is applied to complex industrial scenarios in chemical workshops with high electromagnetic interference, multiple equipment blockages, and strict power consumption control. Through LoRa+WiFi dual-mode adaptive networking and dynamic sleep mechanism, it achieves highly stable and low-power data transmission, as detailed below:

[0062] Dual-mode networking hardware deployment: All edge nodes integrate LoRa low-power communication modules and WiFi high-speed communication modules. The two modules can standby in parallel and switch on demand without manual configuration. It is suitable for complex electromagnetic environments in industrial sites and takes into account the requirements of anti-interference transmission and high-speed data synchronization.

[0063] Communication mode adaptive switching logic: Edge nodes monitor four core parameters in real time: electromagnetic interference intensity, channel signal-to-noise ratio, data packet loss rate, and signal reception strength. A 5% packet loss rate is set as the switching threshold. When the operation of frequency converters or high-power equipment on site causes strong electromagnetic interference, WiFi packet loss rate > 5%, and signal fluctuations are severe, it automatically switches to LoRa anti-interference low-power transmission mode. When the operating conditions are stable, interference is weak, and WiFi signal quality is excellent, it automatically switches to WiFi high-speed transmission mode to quickly synchronize images and full-series data.

[0064] Dynamic sleep-wake low-power mechanism: The system intelligently identifies the device's operating status and divides it into two modes: idle standby and normal operation. When the device is idle standby or there is no dust filtration operation, the sleep mechanism is automatically activated, reducing the data acquisition frequency from once / 1 second to once / 5 minutes and shutting down redundant communication and computing modules. After the device starts running, the full-function modules are automatically woken up, and the high-frequency acquisition, real-time analysis, and high-speed transmission mechanism are restored.

[0065] Data fault-tolerant transmission mechanism: The dual-mode transmission process has built-in data verification, retransmission and packet replenishment mechanisms, which effectively avoid data loss and out-of-order problems caused by industrial obstruction and interference, ensure that each frame of working condition data is uploaded completely and in an orderly manner, and eliminate the problem of misjudgment and omission of working conditions caused by missing data.

[0066] Actual test results: This embodiment has been running for a long time in a high-interference chemical workshop. The average packet loss rate of data in complex scenarios is ≤0.8%, which is significantly lower than the 8.6% packet loss rate of traditional single WiFi networking. The overall power consumption of the equipment is reduced by 63% when the device is idle and by 35% when it is running, which perfectly meets the requirements of long-term low power consumption and high reliability operation in industrial sites.

[0067] Validation data:

[0068] This system was compared with a traditional cloud-based filter cloth monitoring system in the same industrial dust removal site. The test environment was a high electromagnetic interference, multi-obstructed industrial workshop. The system ran continuously for 30 days. The core performance comparison data are as follows: Response latency: The average response latency of this system is 12ms, while the traditional cloud system's average response latency is 2.4s, a 99.5% reduction; Network outage stability: During a 72-hour network outage, this system maintained normal monitoring and early warning without data loss, while the traditional system completely failed; Resource consumption: This system reduced invalid data transmission by 76% and cloud storage usage by 72%; Filter cloth lifespan: After optimizing the cleaning strategy, the average filter cloth lifespan increased by 42.3%, while the traditional timed cleaning mode resulted in a high filter cloth wear rate; Early warning capability: This system provided an average early warning 28 minutes in advance, while the traditional system could only provide alarms after the fact, lacking predictive capability; Transmission performance: The data packet loss rate in complex scenarios was 0.75%, while the traditional single-module network packet loss rate was 8.6%; Equipment power consumption: Idle power consumption was reduced by 62.8%, and operating power consumption was reduced by 35%. The data above demonstrate that this invention has significant performance advantages over traditional technologies in all aspects, and is suitable for stable and efficient operation in complex industrial scenarios.

[0069] This filter cloth condition data acquisition and early warning system based on IoT and edge computing adopts a two-layer architecture of edge-end front-end processing, breaking through the link limitations of traditional pure cloud processing modes. All data processing, condition judgment, and early warning output are completed at the edge nodes in the industrial field, without relying on cloud servers throughout the process. This reduces the traditional cloud-based second-level response latency to millisecond-level, while also possessing complete offline working capabilities. In the event of network anomalies, it can independently complete the entire process of monitoring and early warning, and automatically re-upload synchronized data after reconnecting to the network. This solves the problems of high network dependence, excessive latency, and failure due to network outages in traditional systems, comprehensively ensuring the continuity and stability of industrial filter cloth monitoring. This invention is equipped with a lightweight inference engine, which optimizes the deep neural network model through pruning and quantization techniques, and performs multimodal feature fusion and intelligent condition inference locally. Data upload is triggered only when abnormal conditions are determined; normal condition data is processed locally without uploading, significantly reducing the amount of invalid data transmission, reducing IoT transmission bandwidth usage and cloud storage and computing resource consumption, while avoiding the latency loss caused by round-trip data transmission from the cloud. This achieves millisecond-level accurate identification of six types of filter cloth conditions, improving the real-time performance and resource utilization of condition monitoring. This invention establishes a hybrid decision-making cleaning optimization engine driven by filter cloth degradation types, abandoning the traditional uniform timed cleaning mode. It matches specific control strategies to three core degradation faults: clogging, damage, and cleaning failure, achieving precise intervention and control tailored to each specific problem. This effectively avoids filter cloth mechanical fatigue and fiber peeling damage caused by over-cleaning, as well as filter cloth clogging caused by insufficient cleaning. Actual testing shows that it can effectively extend the service life of the filter cloth and reduce equipment maintenance costs and filter cloth replacement frequency. This invention constructs a lightweight digital twin image of the filter cloth, retaining only core state variables to complete real-time mapping of the three-dimensional state space. Relying on particle filtering algorithms, it accurately predicts the evolution trend of filter cloth resistance and generates fault warning windows in advance. This completely changes the traditional system's passive alarm mode after a fault occurs, achieving a technological leap from passive handling to proactive prediction. The warning lead time is more than three times that of traditional systems, allowing sufficient time for on-site maintenance and equipment control. This invention designs a LoRa+WiFi dual-mode adaptive IoT networking mechanism, which can sense channel interference and network signal quality in industrial sites in real time, automatically switch to the optimal communication mode, and take into account both high-speed transmission and anti-interference requirements. At the same time, it is equipped with a dynamic sleep-wake mechanism to adaptively adjust the data acquisition and transmission frequency, reducing the power consumption of the device in the idle state. It effectively solves the problems of weak anti-interference, high packet loss rate and high power consumption of traditional single-mode networking, and is suitable for complex industrial electromagnetic interference and harsh scenarios with multiple obstructions, ensuring stable data transmission at all times.

[0070] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A filter cloth operating condition data acquisition and early warning system based on the Internet of Things and edge computing, comprising a sensing layer, an edge computing layer, and a cloud management layer, characterized in that: The perception layer includes a multi-dimensional sensor array for collecting multi-modal operating condition data such as filter cloth pressure difference, temperature, instantaneous flow rate, vibration frequency, and surface image. The edge computing layer is deployed in the industrial field and integrates a pre-processing module, a lightweight inference engine, a dust removal strategy optimization engine, a lightweight digital twin mirror, and a dual-mode adaptive networking module. It is used to locally complete data cleaning, time-series feature extraction, intelligent operating condition judgment, precise fault control, operating condition trend prediction, and adaptive data transmission. It supports independent operation offline and automatic data synchronization when connected to the network. This constitutes an edge-front dual-layer processing architecture. The first layer completes the pre-processing and cleaning of raw data, and the second layer completes intelligent decision-making and command output. The cloud-based management layer is used for model iteration, global data storage, and remote device control, enabling millisecond-level real-time monitoring of filter cloth operating conditions, proactive early warning, and precise operation and maintenance control.

2. The filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 1, characterized in that: The lightweight inference engine incorporates a cloud-pruned and quantized one-dimensional deep separable convolutional neural network model, which concatenates multimodal operating condition data into a comprehensive feature vector: ， Where: Xt is the comprehensive feature vector of the filter cloth working condition at time t. denoted as the real-time pressure difference parameter between the front and rear ends of the filter cloth, T as the real-time operating temperature of the filter cloth, Q as the instantaneous flow rate of the filter cloth duct, Vf as the vibration amplitude-frequency feature vector of the filter cloth, and Gt as the surface texture feature vector of the filter cloth at time t.

3. The filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 2, characterized in that: The lightweight inference engine extracts temporal features through one-dimensional depthwise separable convolution operations. The convolution operation formula is as follows: ， Where: Yt is the feature extraction output vector at time t; is a nonlinear activation function; Wd is a depthwise separable convolution weight matrix; * is the convolution operation; b is the convolution bias term, which captures the temporal evolution of filter cloth clogging, scaling, and damage.

4. The filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 3, characterized in that: The lightweight inference engine outputs six types of operational condition confidence scores through a fully connected classification layer, determines anomalies based on preset local alarm thresholds, and when an abnormal operational condition confidence score is reached... The system triggers an early warning, where Si is the real-time confidence score of the i-th type of operating condition, S0 is the preset fixed alarm threshold of the edge node, and i takes values ​​from 1 to 6, corresponding to six types of operating conditions: normal operation, minor blockage, moderate blockage, severe blockage, filter cloth damage, and surface scaling.

5. The filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 4, characterized in that: The lightweight digital twin mirror image uses a particle filter algorithm to predict operating conditions, and the state update formula is as follows: ， in: This is the predicted value of the filter cloth's operating condition at the next moment. The nonlinear evolution transfer function of the filter cloth working condition is given by St, where St is the real-time three-dimensional working condition value of the filter cloth pressure difference, flow velocity, and temperature at time t, and wt is the Gaussian noise parameter of the system process at time t. This function is used to iteratively correct the state estimation error and accurately predict the evolution trend of the filter cloth resistance in the next 30 minutes.

6. The filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 5, characterized in that: The dust removal strategy optimization engine adopts a rule-model hybrid decision-making mechanism to construct a degradation matching formula. ， Wherein: P is the optimal dust removal control parameter set, F is the hybrid decision matching function, Ty is the fault type label, and Kx is the preset dust removal rule parameter library, realizing differentiated and precise control of blockage, damage, and dust removal failure faults.

7. The filter cloth condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 6, characterized in that: The dual-mode adaptive networking module adopts LoRa+WiFi dual communication modes, collects channel interference intensity, network signal-to-noise ratio and packet loss rate parameters in real time, switches to LoRa transmission mode when the interference intensity is higher than a preset threshold, and switches to WiFi high-speed transmission mode when the network signal is good, so as to realize adaptive anti-interference transmission in complex industrial scenarios.

8. The filter cloth condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 7, characterized in that: The pre-processing module is a low-level data processing unit used to receive multimodal raw operating condition data transmitted from the perception layer, and uniformly complete data denoising and filtering, outlier removal, time-series alignment, multi-dimensional data normalization, and image data preprocessing operations. It also normalizes and fuses sensor data of different sampling frequencies and dimensions, removes invalid noise data generated by industrial site interference, and outputs standardized and structured time-series operating condition data, providing a high-quality input data source for the lightweight inference engine and lightweight digital twin mirror.

9. A filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 8, characterized in that: The edge computing layer supports an offline working mechanism when the network is disconnected. When the network is disconnected, it can independently complete data collection, analysis, early warning and local data caching. After the network is restored, it automatically verifies missing data and synchronizes it to the cloud in batches to achieve closed-loop archiving of network data. The edge computing layer is equipped with a dynamic sleep-wake mechanism to identify the device's idle and running states in real time. In the idle state, it automatically reduces the data collection frequency and shuts down redundant transmission modules. In the running state, it restores high-frequency collection and high-speed transmission, and adaptively reduces the device's operating power consumption.

10. A filter cloth working condition data acquisition and early warning system based on the Internet of Things and edge computing according to claim 1, characterized in that: The cloud-based management layer has global data control and model iteration optimization functions, including lightweight model offline training, pruning quantization iteration optimization, remote configuration and distribution of edge node operating parameters, full-scale operating data archiving and storage, historical fault big data tracing and analysis, and equipment operation and maintenance statistical report generation. It also receives abnormal data and offline cached data synchronized from edge nodes, completes global data closed-loop aggregation, and continuously optimizes the parameter library of inference and dust removal strategies of the edge computing layer based on massive cloud data, so as to realize the system's adaptive iterative upgrade.

Images