A fan filter real-time monitoring method based on edge computing

Abstract

The application relates to the technical field of industrial Internet of Things, in particular to a fan filter real-time monitoring method based on edge computing. The method synchronously acquires multi-dimensional data of a fluid field (pressure difference, wind speed), an electrical field (current, harmonic), a mechanical vibration field and an environmental field at a high-frequency sampling rate through an edge computing node arranged at a fan filter device; feature extraction and state reasoning are performed on the edge side, and instantaneous resistance coefficients, mechanical health degrees, electrical health degrees and aerodynamic health degrees are calculated in real time; according to the calculation results, the edge node autonomously performs predictive maintenance early warning and distributed collaborative adjustment based on neighborhood perception. The application realizes local autonomy and predictive maintenance of a fan filter array through calculation sinking, reduces data transmission bandwidth load, improves fault response speed to the millisecond level, and effectively guarantees the stability and safety of an industrial clean environment.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) technology, specifically to a real-time monitoring method for fan filters based on edge computing. Background Technology

[0002] Fan filter units (FFUs) are the cornerstone of modern cleanroom technology. They use their built-in fans to force air through HEPA or UHEPA filters, creating a vertical unidirectional or mixed flow that removes dust particles from the cleanroom and maintains the required cleanliness level for the production environment.

[0003] Traditional FFU monitoring and control systems typically employ a centralized architecture, where all FFUs are connected to one or more central servers via RS485 bus or proprietary protocols. This architecture, when faced with the demands of large-scale, high-precision modern cleanrooms, has gradually revealed a series of insurmountable technical bottlenecks: Data polling latency and poor real-time performance: Traditional systems often use a "master-slave polling" mechanism. The central server needs to query the status of each FFU node in turn. As the number of nodes increases, a complete polling cycle can take several minutes. This means that the system cannot capture transient anomalies, such as voltage drops, short-term wind speed fluctuations, or intermittent mechanical impact sounds.

[0004] The network bandwidth and storage pressures are immense: Early fault detection requires the collection of high-frequency data (such as 10kHz vibration signals or high-sampling-rate current waveforms). In a centralized architecture, uploading all raw data to the cloud or central server would instantly saturate network bandwidth, and the storage and computing pressure on the central server would increase exponentially, potentially causing system crashes or forcing the discard of critical data. Therefore, existing systems typically can only upload low-frequency scalar data (such as a pressure difference value per minute), losing a large amount of high-frequency information rich in fault characteristics.

[0005] The monitoring dimensions are limited, lacking comprehensive diagnostics: Existing systems primarily rely on "differential pressure" and "speed feedback" to determine the FFU status. However, the health of an FFU involves multiple physical fields, including fluid (air path blockage), electrical (coil short circuit, driver aging), and mechanical (bearing wear, dynamic balance failure). Single indicators often have a lag effect. For example, when the differential pressure is significantly abnormal, the filter may already be severely clogged, missing the optimal time for cleaning or replacement; or when the current increases, it may be misinterpreted as increased load when it is actually increased friction due to insufficient lubrication in the motor bearings.

[0006] Lack of local autonomy and coordination capabilities: In centralized control modes, once the network is interrupted or the server crashes, FFUs often can only remain in their last state or shut down, losing their regulatory capabilities. Furthermore, when a faulty FFU causes a "negative pressure hole" in a local area, the central system needs to go through complex global calculations and command issuance processes to schedule surrounding FFUs for compensation, resulting in slow response times. Existing group control systems mostly involve simple start / stop or unified speed adjustment, lacking a distributed intelligent coordination mechanism based on neighborhood awareness.

[0007] To address these challenges, an "edge computing" architecture—which moves computing power from the cloud to the device—has become an inevitable choice. Edge computing allows for real-time processing, analysis, and decision-making at the source of data generation, uploading only high-value information to the cloud. Summary of the Invention

[0008] To address the aforementioned issues, this solution provides a real-time monitoring method for wind turbine filters based on edge computing.

[0009] A real-time monitoring method for wind turbine filters based on edge computing includes the following steps: Data acquisition is performed using edge computing nodes deployed near the physical end of each or each group of wind turbine filters, including: Fluid dynamic parameters: pressure difference between filter inlet and outlet, average air velocity at the outlet cross section; Electrical operating parameters: operating voltage of the drive motor, three-phase or single-phase current; Mechanical vibration parameters: triaxial vibration acceleration of motor bearing housing and fan casing; Environmental parameters: air temperature around the filter; The edge computing node performs real-time processing on the collected raw time-series data locally, including data cleaning and frequency domain conversion; Based on the processed data, a quantitative assessment of health is performed, which integrates features from three dimensions: mechanical health, electrical health, and pneumatic health. The edge computing nodes, based on the results of health quantification assessments, independently or collaboratively execute control strategies, including: proactive hierarchical early warning and distributed neighborhood collaboration; The distributed neighborhood collaboration specifically includes: When an edge node determines that its own fan filter has suffered an irreversible failure and shuts down, it calculates the resulting airflow loss and broadcasts a coordination request via the local area network. After receiving the request, the neighboring edge nodes calculate the incremental airflow they need to bear according to the preset airflow compensation algorithm and automatically increase the output airflow of their respective fan filters to fill the static pressure gap in the fault area.

[0010] Preferably, the method for quantitatively assessing the mechanical health is as follows: Obtain the vibration acceleration sampling sequence within the sampling period; Based on the vibration acceleration sampling sequence, the vibration energy characteristics and kurtosis of the sequence are calculated; Based on the vibration energy characteristics and kurtosis, the vibration offset is calculated: ; In the formula, This is the vibration offset. The preset upper limit of vibration velocity, This is the baseline kurtosis value under a Gaussian distribution. This is the incremental threshold for kurtosis alarm. and These are the preset weighting coefficients; The vibration offset is mapped to the 0-100 range and used as a quantitative assessment result of mechanical health.

[0011] Preferably, the quantitative assessment method for electrical health is as follows: Obtain the high-frequency waveform sampling sequence of the motor stator current; Perform a discrete Fourier transform on the high-frequency waveform sampling sequence to calculate the total harmonic distortion of the current; The winding temperature of the motor's built-in thermistor is obtained, and combined with the ambient temperature obtained from the environmental sensor, the actual temperature rise of the motor is calculated. Based on the total harmonic distortion of the current and the actual temperature rise of the motor, calculate the electrical offset: ; In the formula, This is the electrical offset. This is the preset upper limit threshold for current harmonic distortion. This represents the maximum permissible operating temperature rise for the motor's insulation class. and These are preset weighting coefficients; The electrical offset is mapped to the 0-100 range and used as a quantitative assessment result of electrical health.

[0012] Preferably, the quantitative assessment method for aerodynamic health is as follows: Obtain the initial resistance coefficient of the fan filter during its initial operation. During real-time operation, calculate the current instantaneous drag coefficient; The resistance growth rate is calculated based on the initial resistance coefficient and the current instantaneous resistance coefficient. Get the real-time air volume, and calculate the air volume attenuation rate when the real-time air volume is continuously lower than the set value; Based on the aforementioned resistance growth rate and airflow attenuation rate, calculate the aerodynamic offset: ; In the formula, This is the aerodynamic offset. The maximum allowable resistance increase rate for filter clogging. This represents the percentage of the system's maximum permissible airflow deviation. and These are preset weighting coefficients; The aerodynamic offset is mapped to the 0-100 range and used as a quantitative assessment result of aerodynamic health.

[0013] Preferably, the formula for calculating the instantaneous drag coefficient is: ; In the formula, R(t) is the drag coefficient at time t. Let v(t) be the filter pressure difference at time t, and v(t) be the face wind speed at time t. This is the air density after correction based on real-time temperature and relative humidity.

[0014] Preferably, the air volume compensation algorithm is as follows: ; In the formula, Let $\frac{j}{j}$ be the incremental air volume that the j-th neighboring node needs to handle. This represents the missing airflow value at the faulty node. Let be the distance between the j-th neighbor node and the faulty node. This is the adjustment coefficient.

[0015] Preferably, it also includes establishing a two-way communication mechanism between the edge node and the cloud platform for data uploading. The data uploading strategy specifically includes: normal mode: at fixed intervals, only a JSON format summary package containing the maximum, minimum, average and health scores is uploaded; trigger mode: when the monitoring indicators exceed the safety threshold or the health score undergoes a level jump, the high-frequency raw data before and after the current time window is immediately locked, an event snapshot package is generated and uploaded with the highest priority.

[0016] Compared with the prior art, the advantages of this invention are: Millisecond-level processing capabilities at the edge eliminate the uncertainty of network latency, greatly improving the response speed to rapidly changing faults (such as motor stall).

[0017] It breaks through the limitations of single differential pressure monitoring, integrates vibration, electrical and harmonic analysis, and can identify hidden faults such as early bearing wear and coil insulation aging, truly achieving predictive maintenance.

[0018] The distributed autonomous architecture ensures that even if the central network fails, the fan filter array can still maintain basic constant air volume control; the neighborhood cooperation mechanism enables the system to quickly fill the local static pressure drop caused by the failure of the fan filter using nearby devices, preventing dirty air from flowing back into the clean area from the return air duct, thus improving the stability of the clean environment. Attached Figure Description

[0019] Figure 1 This is a flowchart of a real-time monitoring method for wind turbine filters based on edge computing proposed in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Example 1: A real-time monitoring method for wind turbine filters based on edge computing includes the following steps: Data acquisition is performed using edge computing nodes deployed near the physical end of each or each group of wind turbine filters, specifically including: Fluid sensing: The static pressure difference before and after the HEPA / ULPA filter is monitored by a micro differential pressure sensor, and the air velocity at the outlet surface is monitored by a hot-wire or ultrasonic anemometer.

[0022] Electrical sensing: High-frequency current transformers and voltage probes are used to collect the voltage and current waveforms of the motor power line at high frequency.

[0023] Mechanical sensing: MEMS accelerometers are deployed at key points in the motor bearing housing and the fan volute to collect triaxial vibration signals.

[0024] Environmental sensing: Integrated temperature and humidity sensors are used to compensate for air density calculations, improving the accuracy of air volume calculations.

[0025] Edge computing nodes perform real-time processing of the collected raw time-series data locally, specifically including: Data cleaning: Remove instantaneous noise from sensors by using median filtering or moving average algorithms; synchronize the timestamps of multiple sensors using the IEEE 1588 protocol.

[0026] Frequency domain analysis: Perform FFT transformation on the vibration and current signals in the time domain at the edge to convert the time domain data into frequency domain features (such as fundamental frequency amplitude, harmonic amplitude, and sideband energy).

[0027] Based on the calculated or extracted features, a quantitative assessment of health is performed, specifically including: Based on the collected and processed data, the health status of the machinery is quantified: Edge nodes acquire vibration acceleration sampling sequences within the sampling period and calculate the vibration energy characteristics of the sequences. : ; In the formula, N is the number of samples in the sampling sequence. Let be the vibration acceleration of the i-th sample. When the motor experiences rotor imbalance, It will increase significantly over time.

[0028] Simultaneously calculate the kurtosis K: ; In the formula, Let be the vibration acceleration of the i-th sample. This is the arithmetic mean of the sampled sequences.

[0029] Define an offset function and calculate the normalized vibration offset based on the vibration energy characteristics and kurtosis described above. : ; In the formula, The preset upper limit of vibration velocity, This is the baseline kurtosis value under the Gaussian distribution (taken as 3.0 in this embodiment). This is the incremental threshold for kurtosis alarm. and These are preset weighting coefficients. By setting the weighting coefficients (e.g., ... , The system exhibits higher sensitivity to early impact failures.

[0030] The normalized vibration offset is linearly mapped to a 0-100 score range, which serves as the quantitative assessment result of mechanical health. Edge nodes assess mechanical health every minute. If the health score falls below the mechanical health threshold, the edge node identifies an "early bearing wear" trend and sends preventative maintenance recommendations to the cloud.

[0031] Based on the collected and processed data, the electrical health status is quantitatively assessed: Edge computing nodes acquire high-frequency waveform sampling sequences of motor stator current, and use edge computing power to perform discrete Fourier transform to calculate the total harmonic distortion of the current. : ; In the formula, This is the effective value of the fundamental current. Let be the effective value of the i-th harmonic current, and N be the highest analysis order.

[0032] When the motor magnetic circuit is saturated or electromagnetic interference fluctuations occur in the inverter circuit, the total harmonic distortion of the current will significantly exceed the reference value (e.g., 5%).

[0033] At the same time, the winding temperature of the motor's built-in thermistor is read. And combined with the ambient temperature obtained from environmental sensors Calculate the actual temperature rise of the motor : ; Define an offset function and calculate the normalized electrical offset based on the total harmonic distortion of the current and the actual temperature rise of the motor. : ; In the formula, This is the preset upper limit threshold for current harmonic distortion. This represents the maximum permissible operating temperature rise for the motor's insulation class. and These are preset weighting coefficients. Weighting coefficients (such as...) can be set... , ), to balance the effects of electromagnetic distortion and thermal aging.

[0034] The normalized electrical offset is linearly mapped to a 0-100 score range as the quantitative assessment result of electrical health. Edge nodes assess electrical health every minute. If the health level falls below the electrical health threshold, even if the effective current value is not overloaded, the edge node will identify a "driver modulation stage fault" based on the fluctuation of the electrical health level and send an immediate warning.

[0035] Based on the collected and processed data, the aerodynamic health is quantitatively assessed: Edge computing nodes record the initial resistance coefficient during the initial operation of the wind turbine filter.

[0036] During real-time operation, the current instantaneous resistance coefficient is calculated and compared with the recorded initial resistance coefficient to calculate the resistance growth rate G: ; In the formula, Let be the instantaneous drag coefficient at time t. This represents the initial drag coefficient.

[0037] The formula for calculating the instantaneous drag coefficient is as follows: ; In the formula, R(t) is the drag coefficient at time t. Let v(t) be the filter pressure difference at time t, and v(t) be the face wind speed at time t. This is the air density after correction based on real-time temperature and relative humidity.

[0038] Edge nodes synchronously monitor surface wind speed and calculate real-time airflow. If the real-time airflow remains below the set value due to motor performance degradation or excessive resistance, the airflow attenuation rate A is calculated. ; In the formula, For the pre-set target air volume, This represents the current measured air volume.

[0039] Weight fusion and exponential mapping: Define an offset function and calculate the normalized aerodynamic offset based on the aforementioned drag growth rate and airflow decay rate. : ; In the formula, The maximum allowable resistance increase rate for filter clogging. This represents the percentage of the system's maximum permissible airflow deviation. and These are preset weighting coefficients. Weighting coefficients (such as...) can be set... , 3) Prioritize filter clogging trends.

[0040] The normalized aerodynamic offsets are linearly mapped to a 0-100 score range, serving as the quantitative assessment result of aerodynamic health. Edge nodes perform trend analysis hourly. If the aerodynamic health score falls below the threshold, the edge node calculates the decay slope, predicts the remaining date the index will reach the replacement threshold, and sends a predictive replacement suggestion to the cloud.

[0041] Ultimately, the weighted average or minimum value of the three factors is taken as the overall health score.

[0042] Edge computing nodes, based on their overall health scores, independently or collaboratively execute control strategies, including: Proactive tiered early warning: If the overall health score is greater than 90, it means that the filter of the fan at that node is in a healthy state and no additional action is required.

[0043] If the overall health score is between 75 and 90, it means that the fan filter at that node is in a sub-healthy state. At this time, the edge node predicts the remaining lifespan of the fan filter and suggests the replacement time.

[0044] If the overall health score is below 75, it means that the filter of the fan where the node is located is in a faulty or dangerous state. The edge node will determine the specific fault type (such as "severe bearing wear") and may execute shutdown protection.

[0045] Distributed neighborhood control: Neighbor discovery: Edge nodes identify adjacent nodes in the top, bottom, left, and right directions based on pre-configured physical location IDs and build a neighbor list.

[0046] Collaboration trigger condition: Suppose that the fan filter located at coordinates (x, y) is forced to shut down by the edge node due to a mechanical failure, and its actual output airflow drops to 0. At this time, the node calculates the missing airflow value and sends a collaboration request packet to its neighbor group.

[0047] After receiving the request, the neighboring node calculates the incremental air volume it needs to handle based on a preset compensation algorithm. : ; In the formula, Let $\frac{j}{j}$ be the incremental air volume that the j-th neighboring node needs to handle. This represents the missing airflow value at the faulty node. Let be the distance between the j-th neighbor node and the faulty node. This is the adjustment coefficient.

[0048] The controllers of neighboring nodes adjust their own wind speed setpoints based on the determined incremental air volume they need to handle, and drive the motors to accelerate, thereby filling the static pressure gap in the faulty area. The entire process is completed on the edge network with a latency controlled within 100ms, quickly filling the local static pressure drop caused by the faulty fan filter and preventing dirty air from flowing back into the clean area through the return air duct.

[0049] Example 2: This solution also includes a cloud-edge collaboration mechanism, specifically: Each edge node sends only one JSON-formatted "heartbeat" packet to the cloud per minute, containing data on overall health, mechanical health, electrical health, and pneumatic health. The data size is extremely small (<1KB), and even tens of thousands of devices will not clog the network.

[0050] When an edge node detects that the kurtosis suddenly exceeds a threshold (such as detecting the impact of a bearing breakage), or receives a collaboration request, it triggers "snapshot recording". At this time, the edge node packages the high-frequency vibration waveform data of 5 seconds before and 5 seconds after the failure, marks it as "high priority", and uploads it to cloud object storage.

[0051] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0052] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A real-time monitoring method for wind turbine filters based on edge computing, characterized in that, Includes the following steps: Data acquisition is performed using edge computing nodes deployed near the physical end of each or each group of wind turbine filters, including: Fluid dynamic parameters: pressure difference between filter inlet and outlet, average air velocity at the outlet cross section; Electrical operating parameters: operating voltage of the drive motor, three-phase or single-phase current; Mechanical vibration parameters: triaxial vibration acceleration of motor bearing housing and fan casing; Environmental parameters: air temperature around the filter; The edge computing node performs real-time processing on the collected raw time-series data locally, including data cleaning and frequency domain conversion; Based on the processed data, a quantitative assessment of health is performed, which integrates features from three dimensions: mechanical health, electrical health, and pneumatic health. The edge computing nodes, based on the results of health quantification assessments, independently or collaboratively execute control strategies, including: proactive hierarchical early warning and distributed neighborhood collaboration; The distributed neighborhood collaboration specifically includes: When an edge node determines that its own fan filter has suffered an irreversible failure and shuts down, it calculates the resulting airflow loss and broadcasts a coordination request via the local area network. After receiving the request, the neighboring edge nodes calculate the incremental airflow they need to bear according to the preset airflow compensation algorithm and automatically increase the output airflow of their respective fan filters to fill the static pressure gap in the fault area.

2. The real-time monitoring method for wind turbine filters based on edge computing according to claim 1, characterized in that, The quantitative assessment method for the mechanical health status is as follows: Obtain the vibration acceleration sampling sequence within the sampling period; Based on the vibration acceleration sampling sequence, the vibration energy characteristics and kurtosis of the sequence are calculated; Based on the vibration energy characteristics and kurtosis, the vibration offset is calculated: ； In the formula, This is the vibration offset. The preset upper limit of vibration velocity, This is the baseline kurtosis value under a Gaussian distribution. This is the incremental threshold for kurtosis alarm. and These are the preset weighting coefficients; The vibration offset is mapped to the 0-100 range and used as a quantitative assessment result of mechanical health.

3. The real-time monitoring method for wind turbine filters based on edge computing according to claim 1, characterized in that, The quantitative assessment method for electrical health is as follows: Obtain the high-frequency waveform sampling sequence of the motor stator current; Perform a discrete Fourier transform on the high-frequency waveform sampling sequence to calculate the total harmonic distortion of the current; The winding temperature of the motor's built-in thermistor is obtained, and combined with the ambient temperature obtained from the environmental sensor, the actual temperature rise of the motor is calculated. Based on the total harmonic distortion of the current and the actual temperature rise of the motor, calculate the electrical offset: ； In the formula, This is the electrical offset. This is the preset upper limit threshold for current harmonic distortion. This represents the maximum permissible operating temperature rise for the motor's insulation class. and These are preset weighting coefficients; The electrical offset is mapped to the 0-100 range and used as a quantitative assessment result of electrical health.

4. The real-time monitoring method for wind turbine filters based on edge computing according to claim 1, characterized in that, The quantitative assessment method for aerodynamic health is as follows: Obtain the initial resistance coefficient of the fan filter during its initial operation. During real-time operation, calculate the current instantaneous drag coefficient; The resistance growth rate is calculated based on the initial resistance coefficient and the current instantaneous resistance coefficient. Obtain real-time air volume, and calculate the air volume attenuation rate when the real-time air volume is continuously lower than the set value; Based on the aforementioned resistance growth rate and airflow attenuation rate, calculate the aerodynamic offset: ； In the formula, This is the aerodynamic offset. The maximum allowable resistance increase rate for filter clogging. This represents the percentage of the system's maximum permissible airflow deviation. and These are preset weighting coefficients; The aerodynamic offset is mapped to the 0-100 range and used as a quantitative assessment result of aerodynamic health.

5. The real-time monitoring method for wind turbine filters based on edge computing according to claim 4, characterized in that, The formula for calculating the instantaneous drag coefficient is as follows: ； In the formula, R(t) is the drag coefficient at time t. Let v(t) be the filter pressure difference at time t, and v(t) be the face wind speed at time t. This is the air density after correction based on real-time temperature and relative humidity.

6. The real-time monitoring method for wind turbine filters based on edge computing according to claim 1, characterized in that, The air volume compensation algorithm is as follows: ； In the formula, Let $\frac{j}{j}$ be the incremental air volume that the j-th neighboring node needs to handle. This represents the missing airflow value at the faulty node. Let be the distance between the j-th neighbor node and the faulty node. This is the adjustment coefficient.

7. The real-time monitoring method for wind turbine filters based on edge computing according to claim 1, characterized in that, It also includes establishing a two-way communication mechanism between edge nodes and the cloud platform for data uploading. The specific data uploading strategies include: Normal mode: At fixed intervals, only JSON format summary packages containing the maximum, minimum, average values ​​and health scores are uploaded; Triggered mode: When the monitored indicators exceed the safety threshold or the health score undergoes a level jump, the high-frequency raw data before and after the current time window is immediately locked, an event snapshot package is generated, and uploaded with the highest priority; Resume interrupted transmission: When the network is interrupted, the edge node stores the summary and event data in local non-volatile storage, and retransmits them in timestamp order after the network is restored.

Images