Motor state monitoring method and device based on TinyML

Abstract

The invention relates to the technical field of motor state monitoring, and discloses a TinyML-based motor state monitoring method and device which are applied to a micro-control unit of edge equipment in an HVAC system, and the method comprises the steps: collecting real-time operation parameters of a monitored motor; the real-time operation parameters are preprocessed, the real-time operation parameters are converted into input data in a preset format, and the preset format is a data input format of a TinyML model deployed in a micro-control unit in advance; according to the motor state monitoring method and system, the input data are input into the TinyML model, the real-time state monitoring result of the monitored motor is output through the TinyML model, motor state monitoring is achieved through the micro-control unit of the motor end edge device, the data do not need to be transmitted to a cloud platform in real time for judgment, the monitoring cost is reduced, and the data safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of motor monitoring, and in particular to a motor state monitoring method and device based on TinyML. Background Art

[0002] In residential HVAC systems, the motor is the core component of operation and control. Electronically commutated motors (ECMs) are usually used as starting motors for equipment such as air conditioners.

[0003] In order to monitor the abnormal state of the ECM motor or predict the motor status such as motor life online in real time, some types of machine learning solutions can be deployed by collecting a large amount of data and sending it to a cloud computing platform so that machine learning can be performed. However, this method requires a full-time network connection (such as WIFI, 4G / 5G, etc.) of the electronic commutation motor to transmit data in real time. Due to the need for an additional cloud computing platform and real-time data transmission, this method has the risk of high cost and data being eavesdropped during remote transmission.

[0004] In residential HVAC systems, it is impractical to connect to a cloud-based platform full-time to train a machine learning model, improve it, and then use it to monitor the HVAC system due to high hardware costs and data security issues, not to mention that it is difficult to set up and maintain. Therefore, there is an urgent need for a lightweight, local motor condition monitoring that can determine whether the motor is abnormal or predict the motor life, etc., to solve the high cost and data security issues of existing motor condition monitoring solutions. Summary of the invention

[0005] In view of this, the present invention provides a motor state monitoring method and device based on TinyML to solve the technical problems of high cost and data insecurity of the existing motor state monitoring method based on TinyML.

[0006] In a first aspect, the present invention provides a motor state monitoring method based on TinyML, which is applied to a microcontroller unit of an edge device in an HVAC system, including: collecting real-time operating parameters of a monitored motor; preprocessing the real-time operating parameters, converting the real-time operating parameters into input data in a preset format, wherein the preset format is a data input format of a TinyML model pre-deployed in the microcontroller unit; inputting the input data into the TinyML model, and outputting real-time state monitoring results of the monitored motor through the TinyML model.

[0007] A motor state monitoring method based on TinyML of the present invention collects real-time operating parameters of a monitored motor, pre-processes the real-time operating parameters, converts the real-time operating parameters into input data in a preset format, inputs the input data into the TinyML model pre-deployed in a microcontroller unit, and outputs real-time state monitoring results of the monitored motor through the TinyML model. By utilizing the small scale of the TinyML model and the computing power of the microcontroller unit of the edge device in the HVAC system, the TinyML model is directly deployed on the microcontroller unit of the edge device, and motor state monitoring is implemented through the microcontroller unit of the motor-end edge device. There is no need to transmit data to the cloud platform in real time for judgment, thereby reducing monitoring costs and improving data security.

[0008] Optionally, after outputting the real-time status monitoring result of the monitored motor through the TinyML model, it also includes: adjusting the operating action of the monitored motor according to the real-time status monitoring result.

[0009] In this manner, by adjusting the operation of the motor according to the real-time status monitoring results, the motor operation status can be adjusted in time to prevent the occurrence of failures or accidents.

[0010] Optionally, adjusting the operating action of the monitored motor according to the real-time status monitoring result includes: determining whether the real-time status monitoring result is a first state abnormality; if the real-time status monitoring result is a first state abnormality, changing the speed, start / stop state, or outputting an alarm message of the monitored motor.

[0011] In this way, a quick response can be made to minor abnormal conditions to prevent the fault from worsening by adjusting operating parameters or issuing warnings.

[0012] Optionally, adjusting the operating action of the monitored motor according to the real-time status monitoring result also includes: judging whether the real-time status monitoring result is a second state abnormality; if the real-time status monitoring result is a second state abnormality, sending a maintenance warning message to a third-party platform, the maintenance warning message including the real-time status monitoring result, the identity of the monitored motor and the location information of the monitored motor.

[0013] In this way, when a more serious anomaly occurs, maintenance personnel can be notified in a timely manner, reducing the impact of the fault on system operation and speeding up maintenance response time.

[0014] Optionally, adjusting the operating action of the monitored motor according to the real-time status monitoring result also includes: determining whether the real-time status monitoring result is a third state abnormality; if the real-time status monitoring result is a third state abnormality, sending the collected real-time operating parameters to a third-party platform, and re-determining the state of the monitored motor according to the real-time operating parameters through the third-party platform using a preset artificial intelligence model, wherein the scale or recognition accuracy of the preset artificial intelligence model is greater than that of the TinyML model.

[0015] In this method, for complex or uncertain abnormal conditions, the corresponding real-time operating parameters are transmitted to a third-party platform, and more complex artificial intelligence models are used for in-depth analysis to improve the accuracy of fault diagnosis.

[0016] In this manner, by performing a sliding discrete Fourier transform on the real-time operating parameters, abnormal characteristics in the operation of the motor can be captured more accurately in the frequency domain, thereby improving the sensitivity and accuracy of fault detection.

[0017] Optionally, the TinyML model is a classification model based on a convolutional neural network that has undergone supervised training, the training samples include the corresponding three-phase current in the motor shaft balanced state and the three-phase current in the motor shaft unbalanced state, and the real-time operating parameter is the real-time three-phase current of the monitored motor; correspondingly, the input data is input into the TinyML model, and the real-time status monitoring result of the monitored motor is output through the TinyML model, including: inputting the input data into the TinyML model, and outputting the classification result of the monitored motor through the TinyML model; if the classification result is a shaft balanced state, the output real-time status monitoring result is a normal state, and if the classification result is a shaft unbalanced state, the output real-time status monitoring result is an abnormal state.

[0018] In this method, by taking the three-phase current as input data and classifying the shaft balance and shaft imbalance states of the motor based on the convolutional neural network, the current balance state of the motor can be determined, thereby achieving high-precision fault detection.

[0019] Optionally, the classification model uses a multilayer perceptron, which includes an input layer, a hidden layer and an output layer. The input layer is used to receive the features of the input data, and the hidden layer and the output layer use a nonlinear activation function. The input and output of each hidden layer and the output layer are respectively expressed as:

[0020]

[0021] In the formula, is the input received by the jth neuron in the d+1 layer, n is the number of input features, is the output of the ith neuron in layer d, is the weight of the i-th neuron in layer d on the j-th neuron in layer d+1, represents the threshold of the jth neuron in the d+1 layer, is the input received by the jth neuron in the d layer; the multilayer perceptron uses the directional propagation algorithm to train the weights between neurons in each layer, and the weight update formula is:

[0022]

[0023] In the formula, is the weight in the h layer after update, is the weight in the previous h layers that are updated, μ is the learning rate, and Δ is the difference between the current output of the output layer and the target output.

[0024] In this approach, a multi-layer perceptron is used as the classification model, which is simple to implement and small in scale, making it suitable for MCUs with limited resources, and thus convenient for deployment in microcontroller units of edge devices in HVAC systems.

[0025] In the second aspect, the present invention provides a motor state monitoring device based on TinyML, which is applied to a microcontroller unit of an edge device in an HVAC system, including: a parameter acquisition module, used to acquire real-time operating parameters of a monitored motor; a data preprocessing module, used to preprocess the real-time operating parameters and convert the real-time operating parameters into input data in a preset format, wherein the preset format is the data input format of a TinyML model pre-deployed in the microcontroller unit; and a result detection module, used to input the input data into the TinyML model and output the real-time state monitoring result of the monitored motor through the TinyML model.

[0026] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the TinyML-based motor state monitoring method of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0027] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the TinyML-based motor state monitoring method of the above-mentioned first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 is a flow chart of a motor state monitoring method based on TinyML according to an embodiment of the present invention;

[0030] Figure 2 is a flow chart of another motor state monitoring method based on TinyML according to an embodiment of the present invention;

[0031] Figure 3 is a schematic diagram of a framework of a multilayer perceptron according to an embodiment of the present invention;

[0032] Figure 4 is a schematic diagram showing characteristic data of three-phase current in a shaft-balanced state according to an embodiment of the present invention;

[0033] Figure 5 is a schematic diagram showing characteristic data of three-phase currents in a shaft unbalanced state according to an embodiment of the present invention;

[0034] Figure 6 is a structural block diagram of a motor state monitoring device based on TinyML according to an embodiment of the present invention;

[0035] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] Tiny Machine Learning (TinyML), at the intersection of machine learning and embedded IoT devices, is an emerging engineering discipline with the potential to revolutionize many industries.

[0038] At the same time, the microcontroller unit (MCU) of the electronically commutated motor (ECM) used in the heating, ventilation, and air conditioning (HVAC) system is becoming more and more powerful with the development of technology. Whether it is the same MCU used for motor control or other dedicated MCUs, AI / ML functions can be performed locally without the need for networking. Based on the development of HVAC systems and TinyML models, an embodiment of the present invention proposes a motor state monitoring method based on TinyML, which reduces the cost of motor monitoring and ensures data security by performing machine analysis on edge devices.

[0039] According to an embodiment of the present invention, an embodiment of a motor state monitoring method based on TinyML is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0040] In this embodiment, a motor state monitoring method based on TinyML is provided, which is applied to a microcontroller unit of an edge device in an HVAC system, including a microcontroller unit that controls an electronically commutated motor itself or a microcontroller unit of other dedicated edge devices, such as Figure 1 As shown, the process includes the following steps:

[0041] Step S101, collecting real-time operating parameters of the monitored motor.

[0042] Specifically, the monitored motor is an ECM motor with vector control in an HVAC system, which has a Bluetooth or high-speed optical isolation communication board.

[0043] The real-time operating parameters are collected at a set frequency through the ECM motor's built-in sensors or additionally deployed sensors.

[0044] Real-time operating parameters include speed, torque, power, etc. Multiple real-time operating parameters can be captured synchronously.

[0045] Step S102 , preprocessing the real-time operation parameters, converting the real-time operation parameters into input data in a preset format, wherein the preset format is a data input format of a TinyML model pre-deployed in the micro control unit.

[0046] Specifically, different TinyML models are selected according to different motors and operational failure scenarios in the HVAC field. The function of the TinyML model is mainly to analyze and monitor the motor status, including identifying motor anomalies, failures, or motor life prediction.

[0047] Different types of TinyML models are deployed for different motor and operation fault scenarios. For example, airflow blockage detection and shaft imbalance require very different TinyML models, respectively. In addition, multiple TinyML models can be run in parallel to improve detection accuracy and handle multiple faults at the same time.

[0048] Different types of TinyML models may require different real-time operating parameters and data input formats. For example, in the shaft imbalance anomaly detection scenario, the TinyML model is built based on a convolutional neural network. The real-time operating parameters collected are the three-phase current of the motor. The preprocessing step is mainly Fourier discrete transform. Each abnormal scenario may require a different model or different preprocessing steps for data conversion. In other examples, the preprocessing steps also include abnormal data removal, data normalization, etc.

[0049] Step S103: input the input data into the TinyML model, and output the real-time status monitoring result of the monitored motor through the TinyML model.

[0050] It should be understood that the TinyML model needs to be trained using a pre-collected dataset, preferably supervised training because it may be easier to perform, but other training methods can also be used based on the specific model to be trained. Set up multiple types of TinyML models to train together. After training is completed, a random subset of the collected dataset is used to verify the accuracy of the TinyML model, and the TinyML model with the best prediction accuracy is deployed to the micro-control unit of the edge device in the HVAC system. The TinyML model and motor control code will be combined into deployable firmware and flashed on the micro-control unit of the edge device. Therefore, field testing can be carried out to further verify the efficiency of the model in its natural environment.

[0051] Since the TinyML model deployed in the microcontroller unit has been trained, screened and verified, in actual monitoring, it is only necessary to pre-process the real-time operating parameters collected in real time and convert them into corresponding input data. The real-time status monitoring results can then be directly identified and output through the TinyML model in the microcontroller unit to determine whether the motor has any abnormality or failure.

[0052] A motor state monitoring method based on TinyML in an embodiment of the present invention collects real-time operating parameters of a monitored motor, pre-processes the real-time operating parameters, converts the real-time operating parameters into input data in a preset format, inputs the input data into a TinyML model pre-deployed in a microcontroller unit, and outputs real-time state monitoring results of the monitored motor through the TinyML model. By utilizing the small scale of the TinyML model and the computing power of the microcontroller unit of the edge device in the HVAC system, the TinyML model is directly deployed on the microcontroller unit of the edge device, and motor state monitoring is implemented through the microcontroller unit of the motor-end edge device. There is no need to transmit data to the cloud platform in real time for judgment, thereby reducing monitoring costs and improving data security.

[0053] In some embodiments, after outputting the real-time status monitoring result of the monitored motor through the TinyML model in step S103, the method further includes:

[0054] Step S104: adjusting the operation of the monitored motor according to the real-time status monitoring result.

[0055] Specifically, the use of a microcontroller unit in an edge device to control the monitored motor does not require any additional external equipment, so it can be used for smaller motors (inside a home air conditioner or furnace) and is completely autonomous. This approach has a huge advantage because it allows the MCU to take appropriate actions based on real-time status monitoring results to control motor operation, such as setting an alarm, reducing motor speed, or stopping the motor, depending on the severity of the detected anomaly.

[0056] The severity of the abnormality can be obtained through real-time status monitoring results. For example, the output of the TinyML model includes three results, including normal, first state abnormality and second state abnormality. The severity of the second state abnormality is greater than the severity of the first state abnormality. Furthermore, in order to achieve more precise control of the motor, more subdivision levels of abnormalities can be set.

[0057] In this manner, by adjusting the operation of the motor according to the real-time status monitoring results, the motor operation status can be adjusted in time to prevent the occurrence of failures or accidents.

[0058] In some embodiments, step S104 adjusts the operation of the monitored motor according to the real-time status monitoring result, including:

[0059] Step S1041, determining whether the real-time status monitoring result is that the first status is abnormal.

[0060] Step S1042: if the real-time status monitoring result is that the first status is abnormal, the speed, start / stop status of the monitored motor is changed or an alarm message is output.

[0061] In this embodiment, the deployed TinyML model can at least classify the output real-time status monitoring results into two categories, one is normal and the other is the first state abnormality. When the output result, i.e., the real-time status monitoring result, is the first state abnormality, the speed, start / stop state of the monitored motor is changed or an alarm message is output to prevent the fault from worsening.

[0062] In some embodiments, step S104 adjusts the operation of the monitored motor according to the real-time status monitoring result, and further includes:

[0063] Step S1043, determining whether the real-time status monitoring result is that the second status is abnormal.

[0064] Step S1044: if the real-time status monitoring result is that the second status is abnormal, a maintenance warning message is sent to the third-party platform, where the maintenance warning message includes the real-time status monitoring result, the identity of the monitored motor, and the location information of the monitored motor.

[0065] In this embodiment, the deployed TinyML model can at least classify the output real-time status monitoring results into three categories, one is normal, and the other is the second state abnormality. The severity of the second state abnormality is greater than the severity of the first state abnormality. When the output result, i.e., the real-time status monitoring result, is the second state abnormality, a maintenance warning message is sent to the third-party platform. The maintenance personnel can quickly locate the maintenance problem and the maintenance motor position through the real-time status monitoring result, the identity of the monitored motor and the location information of the monitored motor in the maintenance warning message, so as to facilitate the later maintenance of the motor.

[0066] In this way, when a more serious anomaly occurs, maintenance personnel can be notified in a timely manner, reducing the impact of the fault on system operation and speeding up maintenance response time.

[0067] In some embodiments, step S104 adjusts the operation of the monitored motor according to the real-time status monitoring result, and further includes:

[0068] Step S1045, determining whether the real-time status monitoring result is that the third status is abnormal.

[0069] Step S1046: If the real-time status monitoring result is that the third state is abnormal, the collected real-time operating parameters are sent to the third-party platform, and the third-party platform uses the preset artificial intelligence model to re-judge the state of the monitored motor according to the real-time operating parameters, wherein the scale or recognition accuracy of the preset artificial intelligence model is greater than that of the TinyML model.

[0070] Specifically, the preset artificial intelligence model is a larger-scale and more powerful machine learning model or AI model.

[0071] The third state abnormality is the result of edge device recognition, and the degree of abnormality of the third state abnormality is weaker than that of the first state abnormality. Due to its limited computing resources, it may not be accurate to directly determine that there is an abnormality when the recognition result is the third state abnormality. At this time, by adding a network module or Bluetooth, the real-time operating parameters are transmitted to a third-party platform such as a cloud computing platform in the form of a data stream, so that a more powerful machine learning model or AI model can be used for further and more detailed analysis to obtain more accurate results.

[0072] In this method, for complex or uncertain abnormal conditions, the corresponding real-time operating parameters are transmitted to a third-party platform, and more complex artificial intelligence models are used for in-depth analysis to improve the accuracy of fault diagnosis.

[0073] In some embodiments, step S102, preprocessing the real-time operating parameters to convert the real-time operating parameters into input data in a preset format, includes:

[0074] Step S1021, performing a sliding discrete Fourier transform on the real-time operating parameters to convert the real-time operating parameters into input data in a preset format.

[0075] In this method, the real-time operating parameters are subjected to sliding discrete Fourier transform and converted into input data for the TinyML model. Frequency domain analysis can be performed based on the input data to accurately capture abnormal characteristics in motor operation and improve the sensitivity and accuracy of fault detection.

[0076] In some embodiments, the TinyML model is a classification model based on a convolutional neural network that has undergone supervised training, and the training samples include the three-phase current corresponding to the motor shaft balance state and the three-phase current under the motor shaft unbalance state, and the real-time operating parameter is the real-time three-phase current of the monitored motor. Correspondingly, step S103, inputting the input data into the TinyML model, and outputting the real-time state monitoring result of the monitored motor through the TinyML model, including:

[0077] Step S1031, inputting the input data into the TinyML model, and outputting the classification result of the monitored motor through the TinyML model.

[0078] Step S1032: if the classification result is a shaft balance state, the output real-time state monitoring result is a normal state; if the classification result is a shaft imbalance state, the output real-time state monitoring result is an abnormal state.

[0079] Furthermore, in some embodiments, the classification model uses a multilayer perceptron, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the features of the input data, and the hidden layer and the output layer use a nonlinear activation function. The input and output of each hidden layer and output layer are respectively expressed as:

[0080]

[0081] In the formula, is the input received by the jth neuron in the d+1 layer, n is the number of input features, is the output of the ith neuron in layer d, is the weight of the i-th neuron in layer d on the j-th neuron in layer d+1, represents the threshold of the jth neuron in the d+1 layer, is the input received by the jth neuron in the d layer; the multilayer perceptron uses the directional propagation algorithm to train the weights between neurons in each layer, and the weight update formula is:

[0082]

[0083] In the formula, is the weight in the h layer after update, is the weight in the previous h layers that are updated, μ is the learning rate, and Δ is the difference between the current output of the output layer and the target output.

[0084] Specifically, Multilayer Perceptron (MLP) is a supervised feedforward neural network commonly used in EdgeAI applications. Figure 3 As shown, MLP consists of multiple layers, including an input layer, an output layer, and several hidden layers, each of which contains a group of perceptual elements called neurons. Figure 3 An MLP with two hidden layers, an input layer, and an output layer is shown. In interaction, each node exhibits a certain amount of bias. The input layer contains n input variables X = {x1, x2, …, x n}, the output layer contains m output variables Y = {y1, y2, ..., y m}.

[0085] The hidden layer and the output layer use nonlinear activation functions (sigmoid functions). The total input received by neuron j in the hidden layer of layer d+1 can be declared as:

[0086]

[0087] Except for the input layer, the output of a neuron in any layer can be expressed as:

[0088]

[0089] To train a multilayer perceptron, the backpropagation algorithm should be implemented. This is a supervised learning method because during the learning phase, the output of the multilayer perceptron is compared with a known desired signal to indicate how well the network is performing.

[0090] The backpropagation algorithm is a form of the steepest descent algorithm, i.e., gradient descent, in which the error signal Δ is the difference between the current output of the output layer and the target output before, which is used to adjust the weights in each neuron proportionally to the squared error gradient in the output layer relative to that weight, and then used to adjust the weights in the hidden layer, always returning to the input through the network. Therefore, although the neural network operates on the input signal to give the output in a completely feedforward manner, during the learning process, the error signal generated is propagated from the output of the network back to the input to adjust the weights. That is, the weight update formula for the hidden layer of the hth layer is:

[0091]

[0092] In this approach, a multi-layer perceptron is used as the classification model, which is simple to implement and small in scale, making it suitable for MCUs with limited resources, and thus convenient for deployment in microcontroller units of edge devices in HVAC systems.

[0093] Specifically, Figure 2 As shown, in the embodiment of the present invention, the selected TinyML model is equivalent to a classification model, which divides the three-phase current into an unbalanced shaft state and a balanced shaft state according to the characteristics of the three-phase current. The training samples are obtained by sampling and labeling in advance in a system with a balanced motor shaft and a system with an unbalanced motor shaft.

[0094] Among them, the characteristic data of the three-phase current in the shaft balance state, that is, normal operation, are as follows: Figure 4 As shown in the figure, the characteristic data of the three-phase current in the unbalanced state of the shaft, i.e., in unbalanced operation, are as follows Figure 5 shown.

[0095] The multilayer perceptron is used as the TinyML model, a random subset of the collected training samples is used to verify the trained TinyML model, and the TinyML model with the best prediction accuracy is selected and deployed on the micro control unit of the edge device.

[0096] In one example, the parameters of the selected TinyML model are as follows:

[0097] Score Balance accuracy RAM flash memory Minimum learning iterations Execution time 93.98％ 99.10％ 12.9kb 4.2kb 110 11.8ms

[0098] As can be seen from the above table, when the multilayer perceptron is used as a TinyML model, the required running computing resources and storage resources are relatively small, and it has high accuracy, so it is suitable for deployment on edge devices to achieve edge detection.

[0099] The TinyML model and motor control code will be combined into deployable firmware and flashed on the edge device’s MCU, so field testing can be performed to further validate the model efficiency in its natural environment.

[0100] During real-time monitoring, the real-time sampled three-phase current is subjected to a sliding discrete Fourier transform to obtain input data that conforms to the data format of the TinyML model. The input data is then input into the TinyML model. According to the characteristics of the real-time sampled three-phase current, the monitored motor is divided into a shaft imbalance state and a shaft imbalance state. If the classification result is a shaft balance state, the output real-time state monitoring result is a normal state. If the classification result is a shaft imbalance state, the output real-time state monitoring result is an abnormal state.

[0101] In an embodiment of the present invention, by taking three-phase current as input data and building a TinyML model based on a multilayer perceptron to classify the shaft balance and shaft imbalance states of the motor, the current balance state of the motor can be determined, thereby achieving high-precision fault detection.

[0102] It should be understood that in other embodiments, different input data, different TinyML models, and different data transformations may be used to deploy other scenarios (such as airflow obstruction detection).

[0103] The present invention also provides a motor state monitoring device based on TinyML, which is applied to a microcontroller unit of an edge device in an HVAC system, such as Figure 6 As shown, the device comprises:

[0104] The parameter acquisition module 301 is used to acquire the real-time operating parameters of the monitored motor;

[0105] A data preprocessing module 302 is used to preprocess the real-time operation parameters and convert the real-time operation parameters into input data in a preset format, wherein the preset format is the data input format of the TinyML model pre-deployed in the micro control unit;

[0106] The result detection module 303 is used to input the input data into the TinyML model and output the real-time status monitoring result of the monitored motor through the TinyML model.

[0107] The motor state monitoring device based on TinyML in the embodiment of the present invention collects real-time operating parameters of the monitored motor, pre-processes the real-time operating parameters, converts the real-time operating parameters into input data in a preset format, inputs the input data into a TinyML model pre-deployed in a microcontroller unit, and outputs the real-time state monitoring result of the monitored motor through the TinyML model. By utilizing the small scale of the TinyML model and the computing power of the microcontroller unit of the edge device in the HVAC system, the TinyML model is directly deployed on the microcontroller unit of the edge device, and the motor state monitoring is realized through the microcontroller unit of the motor-end edge device. There is no need to transmit the data to the cloud platform in real time for judgment, thereby reducing the monitoring cost and improving data security.

[0108] Furthermore, the motor condition monitoring device based on TinyML also includes:

[0109] The motion control module is used to adjust the running motion of the monitored motor according to the real-time status monitoring results.

[0110] Furthermore, the action control module includes:

[0111] A first abnormality judgment module, used to judge whether the real-time status monitoring result is a first status abnormality;

[0112] The first control output module is used to change the speed, start / stop state of the monitored motor or output alarm information if the real-time status monitoring result is that the first state is abnormal.

[0113] Furthermore, the action control module includes:

[0114] A second abnormality judgment module is used to judge whether the real-time status monitoring result is a second status abnormality;

[0115] The second control output module is used to send maintenance warning information to the third-party platform if the real-time status monitoring result is that the second state is abnormal. The maintenance warning information includes the real-time status monitoring result, the identity of the monitored motor and the location information of the monitored motor.

[0116] Furthermore, the action control module includes:

[0117] A third abnormality judgment module, used to judge whether the real-time status monitoring result is a third status abnormality;

[0118] The third control output module is used to send the collected real-time operating parameters to the third-party platform if the real-time status monitoring result is that the third state is abnormal, and re-judge the state of the monitored motor according to the real-time operating parameters by using the preset artificial intelligence model through the third-party platform, wherein the scale or recognition accuracy of the preset artificial intelligence model is greater than that of the TinyML model.

[0119] Furthermore, the TinyML model is a classification model based on a convolutional neural network that has undergone supervised training. The training samples include the three-phase current corresponding to the motor shaft balanced state and the three-phase current under the motor shaft unbalanced state. The real-time operating parameter is the real-time three-phase current of the monitored motor.

[0120] Correspondingly, the result detection module 303 includes:

[0121] A data input module is used to input the input data into the TinyML model and output the classification result of the monitored motor through the TinyML model;

[0122] The abnormal confirmation block is used to output the real-time status monitoring result as a normal state if the classification result is a shaft balanced state, and output the real-time status monitoring result as an abnormal state if the classification result is a shaft unbalanced state.

[0123] Furthermore, the classification model uses a multilayer perceptron, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the features of the input data, and the hidden layer and the output layer use a nonlinear activation function. The input and output of each hidden layer and output layer are expressed as:

[0124]

[0125] In the formula, is the input received by the jth neuron in the d+1 layer, n is the number of input features, is the output of the ith neuron in layer d, is the weight of the i-th neuron in layer d on the j-th neuron in layer d+1, represents the threshold of the jth neuron in the d+1 layer, is the input received by the jth neuron in layer d;

[0126] The multilayer perceptron uses the directional propagation algorithm to train the weights between neurons in each layer. The weight update formula is:

[0127]

[0128] In the formula, is the weight in the h layer after update, is the weight in the previous h layers that are updated, μ is the learning rate, and Δ is the difference between the current output of the output layer and the target output.

[0129] The embodiment of the present invention also provides a schematic diagram of the structure of a computer device, such as Figure 7As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 10 is taken as an example.

[0130] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0131] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0132] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0133] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0134] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0135] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0136] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0137] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0138] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope of protection.

Claims

1. A motor state monitoring method based on TinyML, applied to a microcontroller unit of an edge device in an HVAC system, characterized in that: include: Collect the real-time operating parameters of the monitored motor; Preprocessing the real-time operation parameters to convert the real-time operation parameters into input data in a preset format, wherein the preset format is a data input format of a TinyML model pre-deployed in the micro control unit; The input data is input into the TinyML model, and the real-time status monitoring result of the monitored motor is output through the TinyML model.

2. The motor state monitoring method based on TinyML according to claim 1 is characterized in that: After the real-time status monitoring result of the monitored motor is outputted through the TinyML model, the method further includes: The operation of the monitored motor is adjusted according to the real-time status monitoring result.

3. The motor state monitoring method based on TinyML according to claim 2 is characterized in that: Adjusting the operation of the monitored motor according to the real-time status monitoring result includes: Determining whether the real-time status monitoring result is a first status abnormality; If the real-time status monitoring result is that the first state is abnormal, the rotation speed, start / stop state of the monitored motor is changed or an alarm message is output.

4. The motor state monitoring method based on TinyML according to claim 3 is characterized in that: Adjusting the operation of the monitored motor according to the real-time status monitoring result also includes: Determining whether the real-time status monitoring result is that the second status is abnormal; If the real-time status monitoring result is that the second status is abnormal, a maintenance warning message is sent to a third-party platform, where the maintenance warning message includes the real-time status monitoring result, the identity of the monitored motor, and the location information of the monitored motor.

5. The motor state monitoring method based on TinyML according to claim 4 is characterized in that: Adjusting the operation of the monitored motor according to the real-time status monitoring result also includes: Determining whether the real-time status monitoring result is abnormal in the third status; If the real-time status monitoring result is that the third state is abnormal, the collected real-time operating parameters are sent to the third-party platform, and the third-party platform uses a preset artificial intelligence model to re-judge the state of the monitored motor according to the real-time operating parameters, wherein the scale or recognition accuracy of the preset artificial intelligence model is greater than that of the TinyML model.

6. The motor state monitoring method based on TinyML according to claim 1, characterized in that: The TinyML model is a classification model based on a convolutional neural network that has undergone supervised training. The training samples include the three-phase current corresponding to the motor shaft balanced state and the three-phase current under the motor shaft unbalanced state. The real-time operating parameter is the real-time three-phase current of the monitored motor. Correspondingly, the input data is input into the TinyML model, and the real-time status monitoring result of the monitored motor is output through the TinyML model, including: Inputting the input data into the TinyML model, and outputting the classification result of the monitored motor through the TinyML model; If the classification result is a shaft balance state, the output real-time state monitoring result is a normal state; if the classification result is a shaft imbalance state, the output real-time state monitoring result is an abnormal state.

7. The motor state monitoring method based on TinyML according to claim 6 is characterized in that: The classification model adopts a multilayer perceptron, which includes an input layer, a hidden layer and an output layer. The input layer is used to receive the features of the input data, and the hidden layer and the output layer use a nonlinear activation function. The input and output of each hidden layer and the output layer are respectively expressed as: In the formula, is the input received by the jth neuron in the d+1 layer, n is the number of input features, is the output of the ith neuron in layer d, is the weight of the i-th neuron in layer d on the j-th neuron in layer d+1, represents the threshold of the jth neuron in the d+1 layer, is the input received by the jth neuron in layer d; The multilayer perceptron uses the directional propagation algorithm to train the weights between neurons in each layer, and the weight update formula is: In the formula, is the weight in the h layer after update, is the weight in the previous h layers that are updated, μ is the learning rate, and Δ is the difference between the current output of the output layer and the target output.

8. A motor state monitoring device based on TinyML, applied to a microcontroller unit of an edge device in an HVAC system, characterized in that: include: Parameter acquisition module, used to collect real-time operating parameters of the monitored motor; A data preprocessing module, used for preprocessing the real-time operation parameters and converting the real-time operation parameters into input data in a preset format, wherein the preset format is a data input format of a TinyML model pre-deployed in the micro control unit; The result detection module is used to input the input data into the TinyML model, and output the real-time status monitoring result of the monitored motor through the TinyML model.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the motor state monitoring method based on TinyML described in any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the motor state monitoring method based on TinyML according to any one of claims 1 to 7.

Images