A cooling fan system for electric motors

Abstract

This invention provides a cooling fan system for motors, relating to the field of motor cooling technology. The system includes multiple fan casings, a fan, a sensor module, and an intelligent module. The fan casings are open at both ends, and the multiple fan casings are sequentially detachable and connected. The first fan casing's head is connected to the motor end cover, and the fan is located inside the last fan casing. The sensor module acquires basic fan data. The intelligent control unit of the intelligent module controls the fan speed based on the basic data. The isolated forest model branch of the edge fault analysis unit performs anomaly detection on the basic data, generating a basic anomaly score. The lightweight LSTM model branch performs anomaly detection on the basic data and temporal features, generating a temporal anomaly score. The fusion branch weighted and fused the basic anomaly score and the temporal anomaly score to generate a final anomaly score and anomaly detection result. This invention allows for flexible adjustment of airflow distribution.

Description

Technical Field

[0001] This invention relates to the field of motor cooling technology, and more specifically, to a cooling fan system for motors. Background Technology

[0002] Motor cooling is a technical means of dissipating the heat generated by a motor during operation in a timely manner to maintain the motor within a reasonable temperature range. It is a crucial aspect of ensuring motor performance, lifespan, and safety.

[0003] In related technologies, fans used for motor cooling are generally composed of a simple fixed casing and a fan, which cannot meet the needs of different scenarios and cannot reasonably optimize the performance of the fan, thus having certain limitations. Summary of the Invention

[0004] The present invention aims to solve at least one of the above-mentioned problems.

[0005] To address the aforementioned problems, the present invention provides a cooling fan system for an electric motor, comprising multiple fan cylinders, a fan, a sensor module, and an intelligent module. The fan cylinders are open at both ends, and the multiple fan cylinders are sequentially detachable and connected end to end. The first end of the first fan cylinder is used to connect to the motor end cover, and the fan is located inside the last fan cylinder.

[0006] The sensor module is used to acquire basic data of the wind turbine;

[0007] The intelligent module includes an intelligent control unit and an edge fault analysis unit. The intelligent control unit controls the speed of the wind turbine based on the basic data. The edge fault analysis unit includes an isolated forest model branch, a lightweight LSTM model branch, and a fusion branch. The isolated forest model branch performs anomaly detection on the basic data and generates a basic anomaly score. The lightweight LSTM model branch performs anomaly detection on the basic data and the corresponding time-series features and generates a time-series anomaly score. The fusion branch weighted and fused the basic anomaly score and the time-series anomaly score to generate a final anomaly score and an anomaly detection result.

[0008] Optionally, a rear cover plate is connected to the tail end of the last fan cylinder. The rear cover plate has an air inlet. A first air direction filter component and a second air direction filter component are provided between the fan and the rear cover plate. The first air direction filter component is used to adjust the lateral air direction, and the second air direction filter component is used to adjust the longitudinal air direction.

[0009] Optionally, the intelligent module further includes a data uploading unit and a cloud unit;

[0010] The data upload unit is used to determine the fluctuation entropy of the basic data, and when the fluctuation entropy is within a first fluctuation range, upload the anomaly detection result to the cloud unit; when the fluctuation entropy is within a second fluctuation range, upload the anomaly detection result to the cloud unit, and upload the basic data and the time series features to the cloud unit at a preset period; when the fluctuation entropy is within a third fluctuation range, upload the anomaly detection result, the basic data, and the time series features to the cloud unit, wherein the first fluctuation range, the second fluctuation range, and the third fluctuation range increase sequentially;

[0011] The cloud unit is used to store the anomaly detection result when the fluctuation entropy is within the first fluctuation range, and to store the anomaly detection result, the basic data, and the time series features when the fluctuation entropy is within the second or third fluctuation range. It also analyzes the basic data and the time series features and updates the parameters of the lightweight LSTM model branch based on the analysis results.

[0012] Optionally, the cloud unit includes a Transformer model and a knowledge distillation compression layer;

[0013] The Transformer model is trained based on the basic data, the temporal features, and the cross-device correlation features;

[0014] The knowledge distillation compression layer is used to perform knowledge distillation on the trained Transformer model to generate update parameters for the lightweight LSTM model branches.

[0015] The Transformer model includes an input layer, a feature encoding layer, and an anomaly detection layer.

[0016] The input layer is used to input the basic data, the temporal features, and the cross-device association features into the feature encoding layer;

[0017] The feature encoding layer is based on the Transformer's self-attention mechanism to capture the correlation features between the basic data, the temporal features, and the cross-device correlation features;

[0018] The anomaly detection layer is based on a 2-layer fully connected network and outputs the predicted anomaly probability based on the associated features.

[0019] Optionally, the intelligent module further includes a maintenance suggestion unit, which generates maintenance suggestions based on the constructed knowledge graph and using reinforcement learning decision-making according to the objective function.

[0020] Optionally, the intelligent module further includes a dynamic threshold unit, which is used to determine the normal baseline mean and baseline standard deviation of the basic data within a preset time period, and to determine a new threshold based on the normal baseline mean, the baseline standard deviation, a dynamic coefficient, and a dynamic threshold formula, wherein the dynamic threshold formula includes:

[0021] N = P + S × D;

[0022] Wherein, N is the new threshold, P is the normal baseline mean, S is the baseline standard deviation, and D is the dynamic coefficient;

[0023] The fusion branch is also used to generate the anomaly detection result based on the final anomaly score and the new threshold.

[0024] Optionally, the time-series features include statistical features, trend features, and abrupt change features of multiple sliding windows, wherein the statistical features include the mean, variance, and peak-to-trough difference within the window, and the trend features include the linear fitting slope.

[0025] Optionally, an annular groove is provided on the inner side of the first end of the fan casing, and at least one positioning slot is provided on the annular groove. An annular protrusion corresponding to the annular groove is provided on the outer side of the tail end of the fan casing, and at least one positioning block corresponding to the positioning slot is provided on the annular protrusion. The positioning slot on one fan casing is used to engage with the positioning block on another fan casing.

[0026] Optionally, the inner wall of the fan casing is provided with a sound-silencing component, and the last fan casing is provided with an observation window.

[0027] Optionally, a sealing ring is provided at the connection between the first fan casing and the motor end cover.

[0028] The beneficial effects of the cooling fan system for an electric motor according to the present invention are:

[0029] By detachably connecting multiple fan casings end-to-end, the number of fan casings can be set according to actual needs, thereby adjusting the length of the cooling fan system and facilitating adjustments to airflow distribution. This improves the cooling effect on the motor and meets the usage requirements of different scenarios. The detachable fan casings also facilitate maintenance and transportation. Through the intelligent control unit of the intelligent module, the fan speed can be controlled based on the fan's basic data, achieving intelligent control. For example, if the temperature data in the basic data is too high, the fan speed can be appropriately increased to further improve the cooling effect on the motor. Through the isolated forest model branch, lightweight LSTM model branch, and fusion branch of the edge fault analysis unit, basic anomaly detection and temporal anomaly detection are performed on the basic data. Anomaly detection can be performed from both single-point mutation and temporal dimensions, and weighted fusion is used to combine the advantages of both to obtain a final anomaly score. Based on the final anomaly score, anomaly detection results are generated, allowing for reasonable optimization of the fan and improving its performance. Attached Figure Description

[0030] Figure 1 One of the structural schematic diagrams of a cooling fan system for an electric motor provided in an embodiment of the present invention;

[0031] Figure 2 This is a second schematic diagram of a cooling fan system for an electric motor provided in an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of the rear cover plate provided in an embodiment of the present invention;

[0033] Figure 4 One of the schematic diagrams of a first wind direction filter assembly or a second wind direction filter assembly provided in an embodiment of the present invention;

[0034] Figure 5 This is a second schematic diagram of a first or second wind direction filter assembly provided in an embodiment of the present invention.

[0035] Explanation of reference numerals in the attached figures:

[0036] 1. Fan casing; 2. Fan; 3. Rear cover plate; 4. First airflow direction filter assembly; 5. Second airflow direction filter assembly; 6. Noise reduction assembly; 7. Sealing ring. Detailed Implementation

[0037] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0038] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the description below. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0039] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0040] like Figure 1 and Figure 2 As shown in the figure, an embodiment of the present invention provides a cooling fan system for an electric motor, including multiple fan cylinders 1, a fan 2, a sensor module and an intelligent module. The fan cylinders 1 are open at both ends, and the multiple fan cylinders 1 are detachably connected end to end in sequence. The first end of the first fan cylinder 1 is used to connect to the motor end cover, and the fan 2 is located inside the last fan cylinder 1.

[0041] Specifically, the fan casing 1 can be made of high-strength, corrosion-resistant metal materials, such as aluminum alloy or stainless steel, or engineering plastics. The fan casing 1 is a cylindrical structure with open ends. Multiple fan casings 1 can be detachably connected end to end, that is, the tail end of one fan casing 1 can be detachably connected to the head end of the next fan casing 1, for example, by snap-fit ​​or threaded connection. The head end of the first fan casing 1 is used to connect to the motor end cover so that the motor can be cooled by the fan 2 located inside the last fan casing 1. For example, the airflow generated by the fan 2 can blow cold air from the outside to the motor, or the fan 2 can be reversed to carry away the hot air from the surface of the motor through the negative pressure generated by the fan 2. The fan 2 is made of aerospace aluminum alloy or carbon fiber composite material and uses feather-type fan blades to increase the mass, strength and air volume of the fan 2.

[0042] For example, the fan casing 1 can be made of multi-level nested aluminum alloy profiles with built-in guide rails and locking mechanisms. The telescoping ratio is 1:3, and the axial length can be steplessly adjusted between 300 and 900 mm. During transportation, it can be compressed to its shortest size, reducing its volume by 60%, and during installation, it can be dynamically extended according to the motor shaft distance.

[0043] The sensor module is used to acquire basic data of the wind turbine 2.

[0044] Specifically, the sensor module includes various data acquisition devices, such as vibration sensors, temperature sensors, and current sensors. The basic data includes vibration data, temperature data, current data, energy efficiency data, speed data, humidity data, and other data. Among them, vibration data includes amplitude, vibration frequency, and vibration spectrum; temperature data includes the body temperature of fan 2 and ambient temperature; current data includes current value and current frequency; and energy efficiency data includes power and energy consumption.

[0045] The intelligent module includes an intelligent control unit and an edge fault analysis unit. The intelligent control unit is used to control the rotational speed of the wind turbine 2 based on the basic data. The edge fault analysis unit includes an isolated forest model branch, a lightweight LSTM model branch, and a fusion branch. The isolated forest model branch is used to perform anomaly detection on the basic data and generate a basic anomaly score. The lightweight LSTM model branch is used to perform anomaly detection on the basic data and the corresponding time-series features and generate a time-series anomaly score. The fusion branch is used to weightedly fuse the basic anomaly score and the time-series anomaly score to generate a final anomaly score and generate anomaly detection results.

[0046] Specifically, the intelligent module can use a DSP+IPM dual-core chip, integrating Modbus RTU / ASCII dual protocols to realize the functions of each unit, and simultaneously support cluster control of multiple units, such as 128 fans 2. The intelligent control unit is used to control the speed of the fan 2 based on the temperature data in the basic data. For example, when the detected temperature data is less than a first temperature threshold, the speed is controlled to a first preset speed; when the detected temperature data is greater than or equal to the first temperature threshold and less than a second temperature threshold, the speed is controlled to a second preset speed; when the detected temperature data is greater than the second temperature threshold, the speed is controlled to a third preset speed. The first temperature threshold, second temperature threshold, first preset speed, second preset speed, and third preset speed are determined based on actual conditions or experiments to obtain the best cooling effect. For example, the first temperature threshold can be set to 25℃, the second temperature threshold to 45℃, the first preset speed frequency to 40Hz, the second preset speed frequency to 50Hz, and the third preset speed frequency to 60Hz. The edge fault analysis unit can be located at the edge layer of the system and can be mounted on a Raspberry Pi 4B single-board computer. The isolated forest model branch adopts an isolated forest (Isolation)... The Forest model performs anomaly detection on basic data at a single time step, capturing globally sparse outliers (such as single-point mutations and extreme values) to efficiently identify anomalies that deviate significantly from the overall distribution. The lightweight LSTM model branch employs a lightweight LSTM (Tiny-LSTM) model, where each layer has 16 neurons to detect anomalies in the basic data and its corresponding temporal features, capturing temporal pattern anomalies (such as periodic disruptions and gradual drift). The fusion branch calculates a weighted sum to obtain the final anomaly score, which is then compared to a threshold. If the score exceeds the threshold, the anomaly detection result is considered abnormal; otherwise, it is considered normal. The weights of the basic anomaly score and the temporal anomaly score are dynamically adjusted according to the data complexity. For example, low-complexity, stable data assigns a weight of 0.7 to the basic anomaly score, while high-complexity, highly volatile data assigns a weight of 0.8 to the temporal anomaly score. Data complexity can be determined using volatility entropy.

[0047] In this embodiment, multiple fan cylinders 1 are detachably connected end to end, allowing the number of fan cylinders 1 to be set according to actual needs. This adjusts the length of the cooling fan system, facilitating airflow distribution adjustment and improving the cooling effect on the motor to meet the needs of different scenarios. The detachable fan cylinders 1 also facilitate maintenance and transportation. The intelligent control unit of the intelligent module can control the speed of the fan 2 based on its basic data, achieving intelligent control. For example, if the temperature data in the basic data is too high, the speed of the fan 2 can be appropriately increased to further improve the cooling effect on the motor. Through the isolated forest model branch, lightweight LSTM model branch, and fusion branch of the edge fault analysis unit, basic anomaly detection and temporal anomaly detection are performed on the basic data. Anomaly detection can be performed from both single-point mutation and temporal dimensions, and weighted fusion is used to combine the advantages of both to obtain the final anomaly score. Based on the final anomaly score, anomaly detection results are generated, allowing for reasonable optimization of the fan 2 and improving its performance.

[0048] Optionally, such as Figure 1 and Figure 3 As shown, the tail end of the last fan cylinder 1 is connected to a rear cover plate 3. The rear cover plate 3 has an air inlet. A first air direction filter component 4 and a second air direction filter component 5 are provided between the fan 2 and the rear cover plate 3. The first air direction filter component 4 is used to adjust the lateral air direction, and the second air direction filter component 5 is used to adjust the longitudinal air direction.

[0049] Specifically, the tail end of the last fan casing 1 is connected to a rear cover plate 3, which is detachably connected to the fan casing 1. The rear cover plate 3 has openings such as... Figure 3 The air inlet shown includes multiple evenly arranged air inlet holes. A first airflow direction filter assembly 4 and a second airflow direction filter assembly 5 are located between the fan 2 and the rear cover plate 3. The first airflow direction filter assembly 4, used to adjust the lateral airflow, employs a multi-layer filter plate structure. The filter plate material can be stainless steel, glass fiber filter paper, or activated carbon fiber felt, etc., providing high-efficiency filtration performance. The axial projections of the first and second airflow direction filter assemblies 4 and 5 are vertically aligned; that is, the first airflow direction filter assembly 4 is horizontally positioned to adjust the lateral airflow, and the second airflow direction filter assembly 5 is vertically positioned to adjust the longitudinal airflow. Both the first and second airflow direction filter assemblies 4 and 5 include multi-layer filter plates and a screw. The multi-layer filter plates are connected to the screw via a reversing device, thereby changing the angle of the multi-layer filter plates when the screw is rotated. The screw passes through the wall of the last fan cylinder 1 and connects to a knob located on the outside of the last fan cylinder 1. Rotating the screw via the knob changes the angle of the multi-layer filter plates, which can rotate from 0° to 90°. Figure 4As shown, the multi-layer filter plate is rotated to 90°. At this point, the multi-layer filter plate is perpendicular to the axis of the fan casing 1, that is, perpendicular to the wind direction, so as to completely block the wind from the fan 2 and minimize the airflow. Figure 5 As shown, when the multi-layer filter plate is rotated to 0°, it is parallel to the axis of the fan casing 1, that is, parallel to the wind direction, and does not obstruct the airflow of the fan 2, thus maximizing the airflow. It should be noted that... Figure 4 and Figure 5 The two components have the same structure, only their installation orientation is different, so Figure 4 and Figure 5 It can be either a first-direction filter assembly or a second-direction filter assembly. For example, the first-direction filter assembly 4 and the second-direction filter assembly 5 can also be used to filter dust, particles, and other impurities in the airflow, ensuring the purity of the airflow inside the equipment. For example, the knob is mounted on the outer wall of the fan casing 1 via a slide rail groove. The slide rail groove passes through the outer wall of the fan casing 1 to connect the knob and the screw. The slide rail groove is precision-machined, with accurate dimensions and a flat surface, providing stable guidance for the installation of the first-direction filter assembly 4 and the second-direction filter assembly 5. The knob has an adapter slider at its bottom, which can slide within the slide rail groove for easy adjustment and precise control of airflow and filtration effect. The groove wall of the slide rail groove abuts against the outer wall of the screw, ensuring the stability of the first-direction filter assembly 4 and the second-direction filter assembly 5 during the operation of the fan 2, ensuring stable filtration function.

[0050] Optionally, such as Figure 2 As shown, the intelligent module also includes a data uploading unit and a cloud unit;

[0051] The data upload unit is used to determine the fluctuation entropy of the basic data, and when the fluctuation entropy is within a first fluctuation range, upload the anomaly detection result to the cloud unit; when the fluctuation entropy is within a second fluctuation range, upload the anomaly detection result to the cloud unit, and upload the basic data and the time series features to the cloud unit at a preset period; when the fluctuation entropy is within a third fluctuation range, upload the anomaly detection result, the basic data, and the time series features to the cloud unit, wherein the first fluctuation range, the second fluctuation range, and the third fluctuation range increase sequentially;

[0052] The cloud unit is used to store the anomaly detection result when the fluctuation entropy is within the first fluctuation range, and to store the anomaly detection result, the basic data, and the time series features when the fluctuation entropy is within the second or third fluctuation range. It also analyzes the basic data and the time series features and updates the parameters of the lightweight LSTM model branch based on the analysis results.

[0053] Specifically, the data upload unit can be located in the edge layer of the system, and the cloud unit can be located in the cloud layer of the system. The edge layer can communicate with the cloud layer through an OPC UA server to transmit data. The data upload unit determines the fluctuation entropy of the basic data based on the fluctuation entropy formula. When the fluctuation entropy is within the first fluctuation range, it indicates that the data fluctuation is small and relatively stable, with minor problems. Only the anomaly detection results are uploaded to the cloud unit, which greatly reduces the amount of data uploaded. When the fluctuation entropy is within the second fluctuation range, it indicates that the data fluctuation is moderate, with moderate problems. The anomaly detection results are uploaded to the cloud unit, and the basic data and time series features are uploaded to the cloud unit at a preset period, allowing the cloud unit to analyze the basic data and time series features at the preset period. When the fluctuation entropy is within the third fluctuation range, it indicates that the data fluctuation is large and relatively drastic, with significant problems. The anomaly detection results and basic data need to be uploaded to the cloud unit immediately, and the processing priority is marked as "high," so that the cloud unit immediately analyzes the basic data and time series features. The first, second, and third fluctuation ranges are obtained experimentally based on actual conditions and increase sequentially. For example, the first fluctuation range is when the fluctuation entropy is less than 0.3, the second fluctuation range is when the fluctuation entropy is between 0.3 and 0.7, and the third fluctuation range is when the fluctuation entropy is greater than 0.7. The cloud unit can store anomaly detection results, basic data, and temporal features, and analyze the basic data and temporal features. Based on the analysis results, it updates the parameters of the lightweight LSTM model branch, thereby adapting the lightweight LSTM model branch to the existing environment and ensuring detection accuracy. Through the data upload unit and cloud unit of this embodiment, the problem in related technologies that the edge device uploads data at a fixed period regardless of data complexity, and the cloud device passively receives data and can only train periodically, lacking targeted capabilities, can be solved.

[0054] The formula for fluctuation entropy includes:

[0055] ;

[0056] Where E is the fluctuation entropy. p i The probability is the probability that a data point falls into the i-th interval within the sliding window, where n is the number of discrete intervals (the number of bins in the histogram), which determines the granularity of the probability distribution.

[0057] For example, the cloud unit is also used to change the training cycle based on the fluctuation entropy. For instance, when the proportion of high-fluctuation data exceeds 30%, the training cycle is shortened from every 100 data points to every 50 data points; when the proportion of stable data exceeds 70%, the training cycle is extended from every 100 data points to every 200 data points, and the weight of stable data in the training samples is automatically reduced (to avoid overfitting).

[0058] Optionally, the cloud unit includes a Transformer model and a knowledge distillation compression layer;

[0059] The Transformer model is trained based on the basic data, the temporal features, and the cross-device correlation features;

[0060] The knowledge distillation compression layer is used to perform knowledge distillation on the trained Transformer model to generate update parameters for the lightweight LSTM model branches.

[0061] The Transformer model includes an input layer, a feature encoding layer, and an anomaly detection layer.

[0062] The input layer is used to input the basic data, the temporal features, and the cross-device association features into the feature encoding layer;

[0063] The feature encoding layer is based on the Transformer's self-attention mechanism to capture the correlation features between the basic data, the temporal features, and the cross-device correlation features;

[0064] The anomaly detection layer is based on a 2-layer fully connected network and outputs the predicted anomaly probability according to the associated features.

[0065] Specifically, cross-device correlation features refer to concurrent data from similar devices, such as temperature data from other fans, which can be used to verify whether the factors influencing anomalies are environmental. The input layer of the Transformer model is used to input basic data, temporal features, and cross-device correlation features into the feature encoding layer. The feature encoding layer, based on the Transformer's self-attention mechanism, such as a 4-head attention mechanism, captures the correlation features between basic data, temporal features, and cross-device correlation features, such as whether the current is synchronously abnormal when vibration is abnormal. The anomaly detection layer is based on a 2-layer fully connected network, trained according to the correlation features, and outputs the trained anomaly probability. The anomaly probability refers to the probability that the correlation features have an abnormal situation, which may include bearing wear and circuit failure. After training, the Transformer model has a higher accuracy detection capability. The knowledge distillation compression layer is used to perform knowledge distillation on the trained Transformer model to form a lightweight version, thereby reducing the model parameters. The predictions are used as soft labels to train a lightweight LSTM model, generating update parameters for deployable lightweight LSTM model branches.

[0066] Optionally, such as Figure 2 As shown, the intelligent module also includes a maintenance suggestion unit, which generates maintenance suggestions based on the constructed knowledge graph and reinforcement learning decision-making according to the objective function.

[0067] Specifically, a knowledge graph is constructed based on entities and the relationships between them. Entities include equipment models, components (bearings / motors), fault types, maintenance records, and spare parts inventory. Relationships are the associations between different entities. For example, bearing wear is often associated with abnormal vibration and temperature rise, and circuit faults are associated with an average repair time of 2 hours. Reinforcement learning is used for decision-making. Based on an objective function, such as minimizing downtime loss, and a reward function formula, maintenance suggestions are generated. For example, if bearing wear is detected and the remaining lifespan is 5 days, if the current spare parts inventory is sufficient, it is recommended to repair within 2 days; if the spare parts need to be dispatched (arriving in 3 days), it is recommended to repair immediately (to avoid downtime due to stock shortages). For example, the reward function formula includes repair time × production loss coefficient + spare parts transportation cost.

[0068] Optionally, such as Figure 2 As shown, the intelligent module further includes a dynamic threshold unit, which is used to determine the normal baseline mean and baseline standard deviation of the basic data within a preset time period, and to determine a new threshold based on the normal baseline mean, the baseline standard deviation, a dynamic coefficient, and a dynamic threshold formula. The dynamic threshold formula includes:

[0069] N = P + S × D;

[0070] Wherein, N is the new threshold, P is the normal baseline mean, S is the baseline standard deviation, and D is the dynamic coefficient;

[0071] The fusion branch is also used to generate the anomaly detection result based on the final anomaly score and the new threshold.

[0072] Specifically, the dynamic threshold unit determines the normal baseline mean and baseline standard deviation of basic data within a preset time period, such as 7 days. The normal baseline interval is the 95% confidence interval of the non-abnormal data of the previous 7 days. Based on the normal baseline mean, baseline standard deviation, dynamic coefficient, and dynamic threshold formula, a new threshold is determined, which makes the new threshold fluctuate with the baseline. The dynamic coefficient is set according to the equipment's operating years. For example, it is 2.5 when the equipment has been in operation for less than 1 year, 3.0 when it has been in operation for 1 to 3 years, and 3.5 when it has been in operation for more than 3 years. This solves the problem of fixed thresholds in related technologies, which do not take into account factors such as equipment aging (e.g., higher baselines for equipment with longer operating years) and seasonal changes, resulting in inaccurate detection results. The result is an accurate anomaly detection result determined based on the final anomaly score and the new threshold.

[0073] Optionally, the time-series features include statistical features, trend features, and abrupt change features of multiple sliding windows, wherein the statistical features include the mean, variance, and peak-to-trough difference within the window, and the trend features include the linear fitting slope.

[0074] Specifically, the time-series features include statistical features, trend features, and abrupt change features across multiple sliding windows. The number of sliding windows can be experimentally set according to actual conditions; for example, three sliding windows can be used with step sizes of 5 seconds, 10 seconds, and 30 seconds to ensure optimal calculation results. Statistical features include the mean, variance, and peak-to-trough difference within the window. Trend features include the linear fitting slope, which refers to the slope of the time-series data for a single basic data point within the sliding window. For example, for data obtained from vibration, temperature, and current sensors, linear fitting can be performed on continuously acquired data within 5-second, 10-second, and 30-second sliding windows to capture trend features such as "temperature rising within 10 seconds" and "vibration decreasing within 30 seconds," aiding in the identification of latent anomalies. Abrupt change features refer to whether the absolute value of the difference between adjacent data points exceeds twice the historical mean. If it does, the abrupt change feature value is 1; otherwise, it is 0, facilitating model determination of any abrupt anomalies.

[0075] For example, after receiving the anomaly detection results and basic data, the cloud unit is also used to compare the basic data of the same type of wind turbines during the same period. If the temperature of all wind turbines 2 in a certain area rises, the abnormal factor is determined to be an environmental factor, and no maintenance is required for the equipment, thereby reducing the false alarm rate.

[0076] Optionally, an annular groove is provided on the inner side of the first end of the fan cylinder 1, and at least one positioning slot is provided on the annular groove. An annular protrusion corresponding to the annular groove is provided on the outer side of the tail end of the fan cylinder 1, and at least one positioning block corresponding to the positioning slot is provided on the annular protrusion. The positioning slot on one fan cylinder 1 is used to engage with the positioning block on another fan cylinder 1.

[0077] Specifically, an annular groove is provided on the inner side of the first end of the fan cylinder 1, and at least one positioning slot is provided on the annular groove to form a mortise structure. An annular protrusion corresponding to the annular groove is provided on the outer side of the tail end, and at least one positioning block corresponding to the shape of the positioning slot, such as a rectangular block or trapezoidal block, is provided on the annular protrusion to form a tenon structure. The positioning slot and the positioning block are engaged. When the two fan cylinders 1 are connected, the tail end of one fan cylinder 1 is inserted into the head end of the other fan cylinder 1 and rotated to make the positioning block engage in the positioning slot, thereby achieving the engagement effect.

[0078] Optionally, such as Figure 1 As shown, the inner wall of the fan cylinder 1 is provided with a sound-silencing component 6, and the last fan cylinder 1 is provided with an observation window.

[0079] Specifically, the inner wall of the fan casing 1 is equipped with a sound-absorbing component 6. The sound-absorbing component 6 has a honeycomb structure and an array with a diameter of 8 to 12 mm. It can attenuate low-frequency noise of 100 to 500 Hz by 15 dB. The sound-absorbing component 6 can be composed of a combination of sound-absorbing materials (such as sound-absorbing cotton, foam plastic, etc.) and sound-insulating materials (such as rubber sheets, metal sound insulation boards, etc.). The sound-absorbing materials can absorb the noise generated during the operation of the fan and convert the sound energy into heat energy for consumption; the sound-insulating materials can block the outward transmission of noise, effectively reducing the noise pollution generated during the operation of the equipment, and improving the user comfort and environmental friendliness of the equipment. The material of the inner wall of the fan casing 1 is a composite butyl rubber damping layer (thickness of 3 mm) with a loss factor greater than or equal to 0.15, which can suppress the transmission of structural vibration. The last fan casing 1 is equipped with an observation window. The observation window is made of high-strength transparent material (such as plexiglass, tempered glass, etc.), and is connected to the fan casing 1 by a hinge. It is also equipped with a wrench-type opening device to make the opening and closing of the observation window cover simple and quick. Through the observation window, the internal operating status of the equipment can be observed at any time, such as the operation of the fan, the degree of blockage of the first airflow filter component and the second airflow filter component, etc., which facilitates timely maintenance and upkeep.

[0080] For example, the intelligent control unit is also used to adjust the speed of the fan 2 according to the frequency of the motor and the frequency of the fan 2, thereby avoiding frequency resonance between the motor and the fan 2 and reducing resonance noise. In addition, the fan impeller adopts a biomimetic sawtooth tail edge (imitating the structure of an owl feather) to reduce high-frequency eddy current noise.

[0081] Optionally, such as Figure 1 As shown, a sealing ring 7 is provided at the connection between the first fan casing 1 and the motor end cover.

[0082] Specifically, a sealing ring 7 with a thickness of 2.65mm and made of fluororubber is provided at the connection between the first fan casing 1 and the motor end cover. It can withstand temperatures from -20℃ to 200℃.

[0083] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A cooling fan system for an electric machine, characterized by, The application relates to a fan system, which comprises a plurality of fan barrels (1), fans (2), a sensor module and an intelligent module, the first and last ends of the fan barrels (1) are open, the plurality of fan barrels (1) are sequentially and detachably connected, the first end of the first fan barrel (1) is used for being connected with a motor end cover, and the fans (2) are arranged in the interiors of the last fan barrels (1). The sensor module is used for acquiring basic data of the fans (2). The intelligent module comprises an intelligent control unit and an edge fault analysis unit, the intelligent control unit is used for controlling the rotating speed of the fans (2) according to the basic data, the edge fault analysis unit comprises an isolated forest model branch, a lightweight LSTM model branch and a fusion branch, the isolated forest model branch is used for performing abnormal detection on the basic data to generate a basic abnormal score; The lightweight LSTM model branch is used for performing abnormal detection on the basic data and time sequence features corresponding to the basic data to generate a time sequence abnormal score, the fusion branch is used for weightedly fusing the basic abnormal score and the time sequence abnormal score to generate a final abnormal score and an abnormal detection result, wherein the basic data comprises vibration data, temperature data, current data, energy efficiency data, rotating speed data and humidity data, the time sequence features comprise statistical features, trend features and mutation features of a plurality of sliding windows, and the weights of the basic abnormal score and the time sequence abnormal score are dynamically adjusted according to data complexity, and the data complexity is determined by fluctuation entropy.

2. The cooling fan system for an electric machine according to claim 1, characterized by, The last end of the last fan barrel (1) is connected with a rear cover plate (3), the rear cover plate (3) is provided with an air inlet, and first and second air direction filtering assemblies (4) and (5) are arranged between the fans (2) and the rear cover plate (3), the first air direction filtering assembly (4) is used for adjusting a horizontal air direction, and the second air direction filtering assembly (5) is used for adjusting a vertical air direction.

3. The cooling fan system for an electric machine of claim 1, wherein, The intelligent module further comprises a data uploading unit and a cloud unit; The data uploading unit is used for determining the fluctuation entropy of the basic data, and when the fluctuation entropy is in a first fluctuation range, the abnormal detection result is uploaded to the cloud unit; When the fluctuation entropy is in a second fluctuation range, the abnormal detection result is uploaded to the cloud unit, and the basic data and the time sequence features are uploaded to the cloud unit at a preset period; When the fluctuation entropy is in a third fluctuation range, the abnormal detection result, the basic data and the time sequence features are uploaded to the cloud unit, wherein the first fluctuation range, the second fluctuation range and the third fluctuation range increase in sequence; The cloud unit is used for storing the abnormal detection result when the fluctuation entropy is in the first fluctuation range, storing the abnormal detection result, the basic data and the time sequence features when the fluctuation entropy is in the second fluctuation range or the third fluctuation range, and analyzing the basic data and the time sequence features, and updating parameters of the lightweight LSTM model branch according to an analysis result.

4. The cooling fan system for an electric machine according to claim 3, characterized in that, The cloud unit comprises a Transformer model and a knowledge distillation compression layer; The Transformer model is configured to be trained according to the basic data, the time sequence features and the cross-device association features; The knowledge distillation compression layer is configured to perform knowledge distillation on the trained Transformer model to generate updated parameters of the light-weight LSTM model branch. The Transformer model comprises an input layer, a feature encoding layer and an anomaly detection layer. The input layer is configured to input the basic data, the time sequence features and the cross-device association features into the feature encoding layer. The feature encoding layer is based on a self-attention mechanism of the Transformer to capture association features among the basic data, the time sequence features and the cross-device association features. The anomaly detection layer is based on a two-layer fully connected network to output a predicted anomaly probability according to the association features.

5. The cooling fan system for an electric machine of claim 1, wherein, The intelligent module further comprises a maintenance suggestion unit configured to generate a maintenance suggestion according to a target function by using reinforcement learning decision based on the constructed knowledge graph.

6. The cooling fan system for an electric machine of claim 1, wherein, The intelligent module further comprises a dynamic threshold unit configured to determine a normal baseline mean and a baseline standard deviation of the basic data within a preset time, and determine a new threshold according to the normal baseline mean, the baseline standard deviation, a dynamic coefficient and a dynamic threshold formula, the dynamic threshold formula comprising: N = P + S x D; wherein, N is the new threshold, P is the normal baseline mean, S is the baseline standard deviation, and D is the dynamic coefficient. The fusion branch is further configured to generate the anomaly detection result according to the final anomaly score and the new threshold.

7. The cooling fan system for an electric machine of claim 1, wherein, The time sequence features comprise statistical features, trend features and mutation features of a plurality of sliding windows, wherein the statistical features comprise mean, variance and peak-valley difference within a window, and the trend features comprise a linear fitting slope.

8. The cooling fan system for an electric machine of claim 1, wherein, The first end of the fan cylinder (1) is internally provided with an annular groove, and at least one positioning clamping groove is arranged on the annular groove; the tail end of the fan cylinder (1) is externally provided with an annular convex rib corresponding to the annular groove, and at least one positioning clamping block corresponding to the positioning clamping groove is arranged on the annular convex rib; the positioning clamping groove on one fan cylinder (1) is used for clamping the positioning clamping block on another fan cylinder (1).

9. The cooling fan system for an electric machine of claim 1, wherein, The inner wall of the fan cylinder (1) is provided with a mute assembly (6), and the last fan cylinder (1) is provided with an observation window.

10. The cooling fan system for an electric machine of claim 1, wherein, The first fan cylinder (1) is provided with a sealing ring (7) at the connection with the motor end cover.

Images