Overheat protection method for direct-current brush motor

Abstract

The invention discloses a DC brush motor overheating protection method, which relates to the field of motor protection, and comprises the following steps: an electromagnetic wave receiver arranged outside a motor collects electromagnetic radiation signals generated by an electric brush and a commutator in real time when the motor runs; performing multi-frequency band filtering processing on the acquired electromagnetic radiation signal to extract a characteristic frequency component related to overheating, processing the filtered signal by using a dynamic characteristic enhancement algorithm to highlight a waveform mode related to overheating and suppress irrelevant noise, and inputting the processed waveform characteristics into a pre-trained classification model, the overheating risk level of the motor is judged by analyzing the waveform characteristics, load reduction operation or shutdown protection measures are triggered according to the overheating risk level, and when the risk level reaches a preset threshold value, corresponding motor control operation is executed. The problems that a traditional overheating detection method is high in invasiveness and insufficient in accuracy are solved.

Description

Technical Field

[0001] The present invention relates to the field of motor protection, and more particularly, to a method for overheat protection of a DC brushed motor. Background Art

[0002] DC brushed motors are widely used in various fields such as power tools, electric vehicles, and industrial automation due to their simple structure and convenient control. Their working principle is to switch the current direction through the contact between the carbon brush and the commutator, thereby driving the rotor to rotate continuously. However, under high load, long-term operation, or harsh environmental conditions, the friction between the carbon brush and the commutator inside the motor generates a large amount of heat, leading to motor overheating. Overheating not only reduces the performance and lifespan of the motor but also may cause safety hazards such as insulation damage, short circuit, and even fire. Therefore, timely and accurate detection and prevention of motor overheating are the keys to ensuring the safe operation of equipment.

[0003] In the prior art, motor overheat protection usually relies on methods such as temperature sensors or current monitoring. Temperature sensors, such as thermistors, are often installed inside the motor to directly measure temperature changes. However, this method has obvious limitations. In a space-constrained or high-temperature environment, the installation of the sensor not only increases the assembly difficulty but may also fail due to mechanical vibration or high temperature. Another common current monitoring method indirectly infers the heating situation by detecting the input current of the motor, but this method cannot accurately reflect the actual thermal state of the motor and is prone to misjudgment due to load fluctuations or electrical noise. Although these methods alleviate the overheating problem to a certain extent, their deficiencies are still significant, especially the problems of invasive installation and low detection accuracy, which limit their application in compact or complex environments. Therefore, there is an urgent need for a more efficient and non-contact solution to improve the reliability of overheat protection. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for overheat protection of a DC brushed motor to solve the problems mentioned in the background art.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for overheat protection of a DC brushed motor, comprising the following steps:

[0007] An electromagnetic wave receiver arranged outside the motor continuously collects the electromagnetic radiation signals generated by the carbon brush and the commutator during motor operation, and performs multi-frequency band filtering processing on the collected electromagnetic radiation signals to extract the characteristic frequency components related to overheating;

[0008] Use a dynamic feature enhancement algorithm to process the filtered signal to highlight the waveform patterns related to overheating and suppress irrelevant noise, and input the processed waveform features into a pre-trained classification model;

[0009] Judge the overheating risk level of the motor by analyzing the waveform characteristics, trigger the load reduction operation or shutdown protection measures according to the overheating risk level, and execute the corresponding motor control operations when the risk level reaches the preset threshold.

[0010] In an alternative embodiment, the electromagnetic wave receiver is a small wideband antenna, which is installed outside the motor housing and captures the electromagnetic radiation signals generated during the operation of the motor wirelessly;

[0011] The operating frequency range of the antenna is set to 5 kHz to 150 kHz, and the collected signals are converted into digital signals by an analog-to-digital converter and transmitted to the signal processing unit.

[0012] In an alternative embodiment, the multi-band filtering process uses a group of parallel band-pass filters to decompose the collected electromagnetic radiation signals. The frequency range of the band-pass filters is set to 10 kHz to 100 kHz, and the signals are decomposed into three sub-bands: low frequency, medium frequency, and high frequency, where:

[0013] The low-frequency sub-band covers 10 kHz to 30 kHz, the medium-frequency sub-band covers 30 kHz to 70 kHz, and the high-frequency sub-band covers 70 kHz to 100 kHz;

[0014] Generate a frequency feature vector by extracting the amplitude and phase information in each sub-band. The amplitude calculation formula of the feature vector is:

[0015]

[0016] where A(f) is the amplitude at frequency f, R(f) is the real part of the signal, and I(f) is the imaginary part of the signal.

[0017] In an alternative embodiment, the dynamic feature enhancement algorithm uses wavelet transform technology to perform time-frequency analysis on the filtered signal. The signal is decomposed at multiple scales by selecting discrete wavelet basis functions, and the low-order and high-order wavelet coefficients related to overheating are retained. An enhanced waveform feature signal is generated through reconstruction processing, and an amplitude threshold processing is applied to the reconstructed signal to filter out the signal components below the preset amplitude threshold; where the wavelet transform decomposition formula is:

[0018]

[0019] where W(s,t) is the wavelet coefficient, x(τ) is the input signal, ψ * (·) is the conjugate wavelet basis function, s is the scale parameter, t is the time translation parameter, and τ is the integration variable in the formula, which represents the time variable of the input signal x(τ).

[0020] In an alternative embodiment, the pre-trained classification model is a one-dimensional convolutional neural network, and the one-dimensional convolutional neural network is deployed to run in the edge computing module;

[0021] The network structure includes an input layer, a plurality of convolutional layers, a pooling layer, and a fully connected layer connected in sequence. The input layer receives the enhanced waveform feature signal. The convolutional layer extracts the time series features of the waveform through a one-dimensional convolutional kernel. The pooling layer performs dimensionality reduction on the convolutional result. The fully connected layer outputs the classification result of the motor overheating risk. The classification result includes three levels: normal, medium risk, and high risk. The formula for calculating the output features of the convolutional layer is:

[0022] where y i is the i-th output feature, x i+j is the input signal, w j is the convolutional kernel weight, b is the bias, and k is the convolutional kernel length.

[0023] In an alternative embodiment, the protection measures include load reduction operation and shutdown protection. When the overheating risk level output by the classification model is medium risk, the control unit sends an instruction to the motor drive circuit to reduce the input current or speed of the motor. When the overheating risk level output by the classification model is high risk, the control unit immediately cuts off the power supply of the motor to stop the motor operation.

[0024] In an alternative embodiment, the method further includes an online learning mechanism to incrementally update the pre-trained classification model with the electromagnetic radiation waveform data collected in real time; during the operation of the motor, compare the newly collected waveform feature signal with the output result of the classification model. If a classification deviation is detected, add the new data to the training set and adapt to the change of the motor operation state by updating the model weights. The update process is executed at regular time intervals, for example, the time interval can be set to 30 seconds.

[0025] In an alternative embodiment, the online learning mechanism adopts transfer learning technology, uses the weights of the pre-trained classification model as initial parameters, and fine-tunes the model with the waveform data collected in real time. The stochastic gradient descent algorithm is used to optimize the loss function during the fine-tuning process. The loss function is based on the cross-entropy between the output of the classification model and the actual overheating state. For example, the optimization step size can be set to 0.001.

[0026] In an alternative embodiment, the method further includes a multi-source data fusion step. Ambient temperature and motor vibration data are collected by auxiliary sensors integrated in the motor system. The temperature sensor is installed on the surface of the motor housing, and the vibration sensor is installed at the position of the motor bearing. The collected temperature data and vibration data are time-synchronized with the electromagnetic radiation waveform feature signal. The three sets of data are input into a data fusion unit, and a comprehensive overheating risk index is calculated through a weighted average algorithm. For example, the weights can be 0.5 for the electromagnetic radiation waveform feature, 0.3 for the temperature data, and 0.2 for the vibration data.

[0027] In an alternative embodiment, the multi-source data fusion adopts a Bayesian network model. The electromagnetic radiation waveform feature, temperature data, and vibration data are used as input nodes, and the posterior probability of the overheating state is calculated through a pre-constructed conditional probability table. The output of the Bayesian network is a comprehensive overheating risk probability value. A multi-level early warning mechanism is triggered according to the probability value. For example, a low-level warning is issued when the probability value exceeds 50%, and the motor shutdown protection is triggered when the probability value exceeds 80%. The warning signal is transmitted to the motor management system through the control unit.

[0028] The present invention is also applicable to motors with more pole pairs and slots, such as four-pole six-slot brushed motors. Four poles mean there are four magnetic poles. Compared with common two-pole motors, the magnetic field changes more frequently. Six slots mean that each pole will have multiple slot positions to distribute windings. Therefore, the contact changes of the brushes at different slot positions will be more complex. The commutator must handle more current switching events, which makes the electromagnetic radiation signal during the commutation process have more information, and the generated electromagnetic radiation signal is more diverse and has higher spectral characteristics. Therefore, through multi-frequency band filtering and dynamic feature enhancement processing of the electromagnetic radiation signal, the present invention can more accurately extract the characteristic frequency components related to motor overheating, ensuring an accurate judgment of the overheating risk of motors such as four-pole six-slot brushed motors.

[0029] In practical applications, a four-pole six-slot brushed motor is usually controlled by a drive circuit composed of MOSFETs or IGBTs. However, power semiconductor devices (such as MOSFET, IGBT chips and modules) in the drive circuit also generate heat during high-frequency switching, which affects the system stability. Therefore, in alternative embodiments, the technical principle of the present invention can be further extended to power electronic devices in the motor drive system, such as MOSFET and IGBT components. During the high-frequency switching process of such devices, not only characteristic electromagnetic radiation signals are generated, but also abnormal electromagnetic fluctuations may occur due to switching losses, parasitic parameter effects and changes in the operating environment. These abnormal signals are often early signs of increased heat loss, elevated breakdown risk or device aging. If not detected in time, they may lead to faults or even damage to the entire system. The method of the present invention can also be used to analyze the electromagnetic radiation characteristics of these devices through further extension, so as to indirectly evaluate the health status of the drive circuit without adding additional sensors, providing more comprehensive system monitoring capabilities.

[0030] Thus, the advantages of the present invention over the prior art are that by collecting the electromagnetic radiation signals generated during the operation of the motor and performing feature analysis, the overheating state of the motor can be accurately judged. This core innovation effectively solves the problems of insufficient detection accuracy and the need for invasive installation in traditional methods. Its advantages are that no components need to be installed inside the motor, avoiding damage to the equipment caused by high temperature or mechanical stress. At the same time, the contact state between the brush and the commutator is directly reflected through waveform analysis, greatly improving the detection accuracy and response speed. This non-contact method also has strong environmental adaptability and can be widely applied to different types of motors. On this basis, the present invention further enhances the protection effect through a number of auxiliary technologies. Multi-frequency band filtering and dynamic feature enhancement technologies improve the signal quality, ensuring that overheating features can still be accurately extracted under noise interference. The online learning mechanism enables the system to dynamically adjust according to the operating characteristics of a specific motor, reducing false alarms and improving the reliability during long-term use. In addition, the application of multi-source data fusion and Bayesian network integrates multiple information sources and optimizes the risk assessment process. These improvements together construct an efficient and intelligent protection system, which can not only prevent damage caused by overheating in time, but also extend the motor life, reduce maintenance costs, and provide significant technological progress for related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a schematic diagram of the overall method of the present invention;

[0032] Figure 2 is a schematic diagram of the multi-frequency band filtering process of the present invention;

[0033] Figure 3 is a schematic diagram of the dynamic feature enhancement algorithm of the present invention;

[0034] Figure 4 It is a schematic diagram of the online learning of the present invention;

[0035] Figure 5 It is a schematic diagram of the multi-source data fusion mechanism of the present invention. Detailed implementation manners

[0036] The following describes the detailed implementation manners of the present invention with reference to the accompanying drawings.

[0037] The present invention provides an overheat protection method for a four-pole six-slot DC brushed motor. The core lies in collecting the electromagnetic radiation waveform generated during the operation of the motor, analyzing the waveform characteristics in combination with an algorithm, judging whether the motor is in an overheated state, and triggering corresponding protection measures according to the judgment result.

[0038] The method of the present invention is proposed due to the deficiencies of traditional overheat protection technologies in certain scenarios. For example, the methods relying on internal temperature sensors or current monitoring may fail or be difficult to implement in occasions with limited space, high-temperature environments, or high reliability requirements. Therefore, the present invention adopts a non-contact detection means, indirectly judging the overheat situation by using the electromagnetic radiation signal outside the motor. This not only avoids direct intervention in the internal structure of the motor but also improves the applicability and durability of the system. This method provides an innovative and efficient solution for motor overheat protection through the combination of external signal collection and intelligent analysis.

[0039] More specifically, the present invention is based on the following fact: During the operation of the motor, the contact between the brush and the commutator will generate friction and electric sparks, and the intensity and frequency of these phenomena will change with the change of the motor temperature. When the temperature rises, the contact between the brush and the commutator may deteriorate due to thermal expansion or lubricant failure, resulting in an increase in the frequency and intensity of spark discharge. This spark discharge will generate specific electromagnetic radiation signals, and the frequency and amplitude characteristics of which are closely related to the temperature state. Therefore, by collecting these signals, the temperature change of the motor can be indirectly inferred. The paper "Electromagnetic emissions with DC motors–maxon Support" mentions that the electric arc of the brush system is the main source of electromagnetic interference, and the arc intensity may change with the increase of temperature. This provides a theoretical basis for using electromagnetic signals to monitor temperature.

[0040] To ensure that the collected electromagnetic radiation signals can accurately reflect the motor temperature, advanced signal processing techniques can be adopted. For example, through multi-frequency band filtering, the characteristic frequency components related to spark discharge and temperature change are extracted; then, using dynamic feature enhancement algorithms (such as wavelet transform) to highlight the waveform patterns related to temperature while suppressing irrelevant noises (such as environmental noises); finally, combined with a pre-trained classification model (such as a one-dimensional convolutional neural network), the signal features are analyzed to accurately judge the motor temperature state. These technical means can significantly improve the accuracy of temperature monitoring.

[0041] In addition, by conducting experiments on the motor under different temperature conditions, recording the changes in electromagnetic radiation signals, and analyzing their corresponding relationship with temperature, a mapping model between signal features and temperature can be established. In practical applications, the system can accurately infer the motor temperature based on the real-time collected signals in combination with this model.

[0042] More specifically, as Figure 1 shown, in specific implementation, first, the electromagnetic radiation signals generated by the interaction between the brush and the commutator during motor operation are collected in real time through an electromagnetic wave receiver arranged outside the motor. The electromagnetic wave receiver selects a small broadband antenna, which is installed outside the motor housing to capture electromagnetic radiation wirelessly. The operating frequency range of this antenna is set to 5 kHz to 150 kHz, and this range is selected based on a large number of experimental analyses of the motor operating characteristics, covering the typical electromagnetic radiation frequency bands generated when the brush contacts the commutator.

[0043] The collected analog signals are converted into digital signals through an analog-to-digital converter and then transmitted to the signal processing unit for further processing. The design of the small broadband antenna ensures that it can efficiently capture signals under non-contact conditions without interfering with the normal operation of the motor. In practical applications, the antenna is usually placed in a position close to the brush and commutator area to ensure signal strength, and at the same time, the electromagnetic shielding effect of the housing material needs to be considered to avoid signal attenuation.

[0044] The advantage of the broadband antenna is that it can capture various electromagnetic components from low-frequency mechanical vibrations to high-frequency spark discharges, providing comprehensive data support for subsequent feature extraction. For example, when the motor is operating under light load, low-frequency signals may dominate, while when overloaded or overheated, high-frequency spark signals will increase significantly, and this diversity lays the foundation for overheat detection.

[0045] Next, the collected electromagnetic radiation signals need to be processed through multi-frequency band filtering to extract the characteristic frequency components closely related to overheating.

[0046] As Figure 2As shown, this process uses a set of parallel band - pass filters to decompose the signal into multiple sub - bands. The frequency range is set from 10 kHz to 100 kHz, specifically divided into a low - frequency sub - band (10 kHz to 30 kHz), a medium - frequency sub - band (30 kHz to 70 kHz), and a high - frequency sub - band (70 kHz to 100 kHz). By analyzing the amplitude and phase information in each sub - band, a frequency feature vector is generated, and its amplitude calculation formula is:

[0047]

[0048] where \(A(f)\) is the amplitude at frequency \(f\), \(R(f)\) is the real part of the signal, and \(I(f)\) is the imaginary part of the signal.

[0049] The motivation for this decomposition method is that when the motor overheats, the contact state between the brush and the commutator will change. For example, the spark intensifies or the noise increases, and these changes are particularly obvious within a specific frequency range. By splitting the signal into different sub - bands, these features can be captured more precisely.

[0050] For example, the low - frequency sub - band may reflect the cumulative effect of mechanical vibration, the medium - frequency sub - band may be related to the electrical property changes when the heat slowly rises, while the high - frequency sub - band is more likely to capture the rapid discharge phenomenon caused by overheating. In actual operation, the bandwidth and frequency range of the filter can be fine - tuned according to the motor model, load conditions, or operating environment to improve the pertinence and accuracy of detection. This flexibility ensures the adaptability of the method in different application scenarios.

[0051] Subsequently, the filtered signal is further processed by a dynamic feature enhancement algorithm to highlight the waveform patterns related to overheating while suppressing the irrelevant background noise.

[0052] As Figure 3 shown, this algorithm uses wavelet transform technology to perform time - frequency analysis on the signal, realizes multi - scale decomposition through discrete wavelet basis functions, and retains the low - order and high - order wavelet coefficients related to overheating. The formula is:

[0053]

[0054] where \(W(s,t)\) is the wavelet coefficient, \(x(τ)\) is the input signal, \(\psi\) * (·) is the conjugate wavelet basis function, \(s\) is the scale parameter, and \(t\) is the time translation parameter.

[0055] After decomposition, an enhanced waveform feature signal is generated through reconstruction, and an amplitude threshold processing is applied to the reconstructed signal to filter out components below the preset threshold. The advantage of wavelet transform lies in its ability to analyze both the time-local characteristics and frequency characteristics of the signal simultaneously, especially suitable for processing non-stationary signals such as motor electromagnetic radiation. For example, when the motor generates sudden spark discharges due to overheating, high-frequency mutations will appear in the signal, and wavelet transform can effectively capture these characteristics. Selecting the Daubechies wavelet as the basis function can better extract such mutations or high-frequency components, and at the same time eliminate irrelevant noise through threshold processing, improving the signal-to-noise ratio of the signal. The implementation of this processing not only improves the accuracy of feature extraction but also lays a solid foundation for subsequent classification.

[0056] The processed waveform feature signal is input into a pre-trained classification model to analyze and judge the overheating risk level of the motor. This model uses a one-dimensional convolutional neural network and is deployed to run in an edge computing module. Its network structure includes an input layer, multiple convolutional layers, pooling layers, and fully connected layers connected in sequence. The input layer receives the enhanced waveform feature signal. The convolutional layer extracts time series features through a one-dimensional convolutional kernel. The pooling layer reduces the dimension of the convolutional result. The fully connected layer finally outputs the classification result of the overheating risk, which is divided into three levels: normal, medium risk, and high risk. The calculation formula for the features output by the convolutional layer is

[0057]

[0058] where y i is the i-th output feature, x i+j is the input signal, w j is the convolutional kernel weight, b is the bias, and k is the length of the convolutional kernel.

[0059] The original intention of this design is to use deep learning to automatically learn the complex relationship between waveform features and overheating states. Compared with traditional threshold judgment, its accuracy and robustness are higher. Deploying the model in the edge computing module realizes real-time analysis and quick response, avoiding the delay of data transmission to the cloud. For example, when the motor runs at high load for a long time, the model can identify the gradually increasing risk trend through the training of historical waveform samples. The model training needs to use waveform data covering various operating conditions to ensure its generalization ability in different scenarios.

[0060] According to the overheating risk level output by the classification model, the system will trigger corresponding protection measures. When the classification model determines that the motor is at medium risk, the system sends an instruction to the motor drive circuit through the control unit to reduce the input current or speed of the motor. The realization of this measure depends on the adjustability of the motor drive circuit. For example, the input voltage is adjusted through Pulse Width Modulation (PWM) technology to reduce the current input. Reducing the current or speed can effectively reduce the heat generation of the motor and enable it to cool down gradually without overheating.

[0061] For example, in an industrial assembly line, the motor may need to run for a long time. At this time, reducing the current by 20% or the speed by 15% can alleviate the overheating trend without completely interrupting production. The specific load reduction amplitude can be preset according to the rated power of the motor and the current load situation, and optimized through experiments in actual applications.

[0062] When the risk level rises to high, the system will immediately cut off the power supply of the motor and trigger shutdown protection. This is achieved through a hardware interruption mechanism. The control unit directly sends a shutdown signal to the power relay to ensure that the response time is controlled within the millisecond level, such as less than 10 milliseconds, to prevent insulation damage or fire risk caused by overheating. The motivation for this hierarchical protection strategy is to balance operation continuity and safety, avoid unnecessary shutdowns, and take decisive actions in high-risk situations. In electric vehicle applications, this mechanism can protect key components when the motor is overloaded and extend the service life.

[0063] As Figure 4 shown, an online learning mechanism can be further adopted to update the classification model through the real-time collected electromagnetic radiation waveform data to adapt to the changes of specific motors and operating environments. The initial classification model is pre-trained based on a general dataset. However, due to the individual differences of motors (such as manufacturing tolerances or aging) and the changes of external conditions (such as environmental temperature or load fluctuations), the model may gradually deviate from the actual state. The implementation process of online learning includes regularly collecting the electromagnetic radiation signals during motor operation and comparing them with the actual overheating state.

[0064] For example, the system collects data every 30 seconds, inputs the waveform features into the classification model, and calculates the deviation based on the detection result (whether it is truly overheated). If a prediction error is found, new data is added to the training set and the model parameters are adjusted. This process is completed through incremental learning, avoiding retraining all the data and reducing the computational burden. In an embedded system, the update frequency and data volume need to be adjusted according to the processor performance. For example, on a microcontroller with medium performance, the number of updates per day can be limited to within 50 times. The advantage of this mechanism is that it enables the system to dynamically adapt to the personalized characteristics of the motor. For example, a motor that has been used for many years may exhibit different electromagnetic radiation patterns due to wear. Online learning can capture these changes and optimize the model, thereby improving the detection accuracy and reducing false alarms or missed detections.

[0065] In a further embodiment, online learning adopts transfer learning and Stochastic Gradient Descent (SGD) algorithm. The core idea of transfer learning is to use the general knowledge of the pre-trained model as a starting point and then fine-tune it with real-time data. The pre-trained model may be trained based on the electromagnetic radiation data of a large number of standard motors and contains general features related to overheating. In practical applications, the system inputs the waveform features collected in real time into the model, calculates the loss (such as cross-entropy loss) between the prediction result and the actual state, and updates the model weights through SGD. The advantage of SGD lies in its incremental update characteristic. Each time, only a small amount of new data (such as a batch contains 10 samples) is used to calculate the gradient, which is suitable for scenarios with high real-time requirements. The learning rate is set to 0.001. This relatively small value ensures smooth weight adjustment and avoids model divergence caused by too large a step size.

[0066] For example, on a newly installed motor, the system may quickly adapt to its characteristics through transfer learning within the first few hours of operation, significantly reducing misjudgments in the initial stage. Compared with training from scratch, transfer learning requires less data volume. Usually, only hundreds of samples are needed to achieve high accuracy. This method is particularly suitable for scenarios such as motor replacement or changes in the operating environment, such as mobile devices operating under different climate conditions, and can quickly adjust the model to maintain reliability.

[0067] As Figure 5 shown, in a further embodiment, a multi-source data fusion mechanism can be adopted. The specific method is to integrate the electromagnetic radiation waveform features with temperature and vibration data. The temperature sensor is installed on the motor housing to monitor the surface temperature in real time; the vibration sensor is placed at the bearing to detect mechanical vibrations during operation. These two types of data are synchronized with the electromagnetic radiation signal in time to form a multi-modal data set.

[0068] The fusion can adopt the weighted average algorithm, and the weights of each data source are determined according to their correlation with overheating. For example, the weight of the electromagnetic radiation waveform feature is set to 0.5, the temperature data is 0.3, and the vibration data is 0.2. The process of data fusion includes normalizing each sensor signal (for example, normalizing it to the range of 0 - 1), and then calculating the weighted sum as the comprehensive risk index. In practical applications, if the temperature suddenly rises but the electromagnetic radiation signal is normal, the system may determine it as an external environmental impact rather than motor overheating, thus avoiding misjudgment. The motivation for multi-source fusion is that a single signal may fail due to noise or anomalies, while multi-modal data can verify each other and improve robustness. For example, during heavy load operation, the vibration data may indicate bearing problems, and the temperature and electromagnetic radiation signals jointly confirm the overheating risk, thus triggering protective measures. The specific values of the weights can be optimized through historical data analysis or machine learning to ensure the optimal fusion result.

[0069] In a further embodiment, the data fusion can adopt a Bayesian network and combine a multi-level early warning mechanism for risk management. The Bayesian network is a probabilistic graphical model, which includes nodes such as electromagnetic radiation characteristics, temperature, vibration, and overheating status. The conditional dependence relationships between the nodes are defined by a preset conditional probability table (CPT). For example, the CPT may indicate that when the electromagnetic radiation signal is abnormal and the temperature is higher than 80 °C, the probability of overheating is 90%. The system inputs real-time data into the network and calculates the posterior probability of the overheating status through the probability propagation algorithm. This probability value drives the multi-level early warning mechanism: if the probability exceeds 50%, a low-level warning is triggered, for example, by flashing an LED light or sounding a buzzer to alert the operator; if the probability exceeds 80%, a high-level warning is triggered and the machine is immediately stopped.

[0070] The threshold setting is based on the actual needs of risk management. 50% provides an opportunity for early intervention, and 80% ensures timely protection in high-risk states. In an actual system, the inference process of the Bayesian network can be completed within a few milliseconds, meeting the real-time requirement. The construction of the conditional probability table can combine historical data and expert knowledge. For example, the correlation between variables is determined by analyzing 1000 hours of operation data. Compared with the simple threshold method, the Bayesian network can quantify uncertainty and provide a more accurate risk assessment. For example, in a scenario with a high ambient temperature, the network can avoid false alarms triggered by a single temperature increase through comprehensive analysis. The advantage of this method lies in its flexibility and reliability, which is suitable for motor protection under complex operating conditions.

[0071] The present invention is also applicable to motors such as four-pole six-slot brushed motors with more pole pairs and slots. Four poles indicate that there are four magnetic poles. Compared with common two-pole motors, the magnetic field changes more frequently. Six slots mean that there are multiple slot positions for each pole to distribute windings. Therefore, the contact changes of the brush at different slot positions will be more complex. The commutator must handle more current switching events, which makes the electromagnetic radiation signals during the commutation process have more information, and the generated electromagnetic radiation signals are more diverse and have higher spectral characteristics. Therefore, through multi-frequency band filtering and dynamic feature enhancement processing of the electromagnetic radiation signals, the present invention can more accurately extract the characteristic frequency components related to motor overheating, ensuring an accurate judgment of the overheating risk of motors such as four-pole six-slot brushed motors.

[0072] In summary, through non-contact electromagnetic radiation signal acquisition, combined with multi-frequency band filtering, wavelet transform enhancement, and convolutional neural network classification, the present invention realizes precise monitoring and timely protection of the overheating state of four-pole six-slot DC brushed motors. This method not only overcomes the limitations of traditional technologies but also provides innovative ideas for the field of motor protection and has significant application potential. The implementation of further hierarchical protection measures, dynamic adaptation of online learning, comprehensive analysis of multi-source data fusion, and probabilistic decision-making of Bayesian networks jointly construct an efficient, flexible, and reliable overheating protection system. Through load reduction and shutdown protection, the system can effectively respond under different risk levels; online learning and transfer learning ensure the adaptability of the model to individual motors and environmental changes; multi-source fusion and Bayesian networks significantly improve the accuracy and robustness of detection. In practical applications, such as industrial motors or electric vehicle motors, the system can extend the equipment life, reduce maintenance costs, and ensure operation safety, with significant practical value and promotion potential.

[0073] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for overheat protection of a DC brushed motor, characterized in that It includes the following steps: An electromagnetic wave receiver arranged outside the motor collects in real time the electromagnetic radiation signals generated by the brush and the commutator during the operation of the motor, and performs multi-frequency band filtering processing on the collected electromagnetic radiation signals to extract the characteristic frequency components related to overheating; The filtered signal is processed using a dynamic feature enhancement algorithm to highlight the waveform patterns related to overheating and suppress irrelevant noises, and the processed waveform features are input into a pre-trained classification model; The overheating risk level of the motor is judged by analyzing the waveform features, and the load reduction operation or shutdown protection measures are triggered according to the overheating risk level. When the risk level reaches the preset threshold, the corresponding motor control operations are executed.

2. The overheat protection method for a DC brushed motor according to claim 1, wherein The electromagnetic wave receiver is a small wide-band antenna, which is installed outside the motor housing and captures the electromagnetic radiation signals generated during the operation of the motor wirelessly; The operating frequency range of the antenna is set to 5 kHz to 150 kHz, and the collected signals are converted into digital signals by an analog-to-digital converter and transmitted to the signal processing unit.

3. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that: The multi-frequency band filtering processing uses a group of parallel band-pass filters to decompose the collected electromagnetic radiation signals. The frequency range of the band-pass filters is set to 10 kHz to 100 kHz, and the signals are decomposed into three sub-bands: low frequency, medium frequency, and high frequency, where: The low-frequency sub-band covers 10 kHz to 30 kHz, the medium-frequency sub-band covers 30 kHz to 70 kHz, and the high-frequency sub-band covers 70 kHz to 100 kHz; Frequency feature vectors are generated by extracting the amplitude and phase information in each sub-band. The amplitude calculation formula of the feature vector is: where A(f) is the amplitude at frequency f, R(f) is the real part of the signal, and I(f) is the imaginary part of the signal.

4. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that, The dynamic feature enhancement algorithm uses wavelet transform technology to perform time-frequency analysis on the filtered signal. The signal is decomposed at multiple scales by selecting discrete wavelet basis functions, and the low-order and high-order wavelet coefficients related to overheating are retained. An enhanced waveform feature signal is generated through reconstruction processing, and an amplitude threshold processing is applied to the reconstructed signal to filter out the signal components below the preset amplitude threshold; where the wavelet transform decomposition formula is: Among them, W(s,t) is the wavelet coefficient, x(τ) is the input signal, and ψ * (·) is the conjugate wavelet basis function, s is the scale parameter, t is the time translation parameter, and τ is the integration variable in the formula, which represents the time variable of the input signal x(τ).

5. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that, The pre-trained classification model is a one-dimensional convolutional neural network, and the one-dimensional convolutional neural network is deployed to run in the edge computing module; The network structure includes an input layer, multiple convolutional layers, a pooling layer, and a fully connected layer connected in sequence. The input layer receives the enhanced waveform feature signal. The convolutional layer extracts the time series features of the waveform through a one-dimensional convolutional kernel. The pooling layer performs dimensionality reduction on the convolutional result. The fully connected layer outputs the classification result of the motor overheating risk. The classification result includes three levels: normal, medium risk, and high risk. The formula for calculating the output features of the convolutional layer is: where y i is the i-th output feature, x i+j is the input signal, w j is the convolutional kernel weight, b is the bias, and k is the convolutional kernel length.

6. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that, The protection measures include load reduction operation and shutdown protection. When the overheating risk level output by the classification model is a medium risk, an instruction is sent to the motor drive circuit through the control unit to reduce the input current or speed of the motor. When the overheating risk level output by the classification model is a high risk, the control unit immediately cuts off the power supply of the motor to stop the motor operation.

7. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that, The method also includes an online learning mechanism to incrementally update the pre-trained classification model through the electromagnetic radiation waveform data collected in real time; during the operation of the motor, the newly collected waveform feature signals are compared with the output results of the classification model. If a classification deviation is detected, the new data is added to the training set, and the model weights are updated to adapt to the changes in the motor operation state. The update process is executed at fixed time intervals.

8. A method for overheat protection of a DC brushed motor according to claim 7, characterized in that, The online learning mechanism adopts transfer learning technology, uses the weights of the pre-trained classification model as initial parameters, and fine-tunes the model using the waveform data collected in real time. During the fine-tuning process, the stochastic gradient descent algorithm is used to optimize the loss function, and the loss function is based on the cross-entropy between the output of the classification model and the actual overheating state.

9. A method for overheat protection of a DC brushed motor according to claim 1, characterized in that The method further includes a multi-source data fusion step. Ambient temperature and motor vibration data are collected through auxiliary sensors integrated in the motor system. The temperature sensor is installed on the surface of the motor housing, and the vibration sensor is installed at the motor bearing position. The collected temperature data and vibration data are time-synchronized with the electromagnetic radiation waveform feature signal, and the three groups of data are input into the data fusion unit, and the comprehensive overheating risk index is calculated through the weighted average algorithm.

10. A method for overheat protection of a DC brushed motor according to claim 9, characterized in that, The multi-source data fusion adopts a Bayesian network model. The electromagnetic radiation waveform features, temperature data, and vibration data are used as input nodes, and the posterior probability of the overheating state is calculated through a pre-constructed conditional probability table. The output of the Bayesian network is the comprehensive overheating risk probability value. According to the probability value, a multi-level early warning mechanism is triggered, and the warning signal is transmitted to the motor management system through the control unit.

Images