A detection method of an angular position sensor

Abstract

A detection method of an angle position sensor, when the angle position sensor fails, according to the threshold value established by the convolutional neural network and the long short-term memory neural network, it is judged whether a single sensor fails or both sensors fail, the angle position sensor can be replaced in time, the data error in detection is avoided, and the final detection result is ensured.

Description

Technical Field

[0001] This invention relates to the field of angular position sensor technology, and more specifically to a method for detecting angular position sensors. Background Technology

[0002] Angular position sensors are mainly used in missile guidance and control, artillery servo systems, radar antenna tracking systems, aircraft attitude control systems, robotic systems, electric vehicles, and other fields. However, there is a lack of protective measures for the sensors and monitoring and diagnosis of their working status. When angular position sensors malfunction, the data detected by the sensors is inaccurate, affecting the final detection results and causing fatal judgment errors, which may result in huge losses.

[0003] The existing method is to set up redundant sensors and use two sensors to determine whether one of them has failed. However, when both sensors fail, it is impossible to determine whether a single sensor or both sensors have failed. When a single sensor fails, it is impossible to determine which sensor has failed. Summary of the Invention

[0004] To address the shortcomings of the prior art, this invention provides a method for detecting angular position sensors that uses a convolutional neural network and a long short-term memory neural network to determine whether an angular position sensor has failed.

[0005] The technical solution of the present invention is: a detection method for an angular position sensor, comprising the following steps:

[0006] S1: Data modeling is performed using convolutional neural networks and long short-term memory neural networks based on the data recorded by the corner position sensors, and two corner position sensors are set up for detection;

[0007] S2: Based on the modeling data, detect whether the output values ​​of the two corner position sensors are maintained within the range and whether they are always fixed values ​​to determine whether the two corner position sensors are malfunctioning;

[0008] S3: Compare the predicted value of the modeling data with the output values ​​of the two corner position sensors. If the output value of one sensor matches its corresponding predicted value, while the output value of the other sensor differs significantly from its corresponding predicted value, the system determines that one of the corner position sensors has failed.

[0009] S4: When the output values ​​of both angular position sensors differ significantly from the predicted values, the error is calculated using the following formula:

[0010]

[0011] in the formula The value represents the average of the first and second angle position sensors, f represents the error caused by the angle position sensor malfunction, and t represents time.

[0012] S5: Determine whether the angular position sensor is faulty based on the change in the value of f. If the initial value of f jumps and the subsequent value of f returns to a stable state, i.e., σ(f) < γ, where γ is the error reference value, then the two angular position sensors are not faulty. Retrain the neural network based on the output value of the angular position sensor when the working conditions change drastically. After steps S3 and S4, if the value of f does not jump or the subsequent σ(f) > γ, then the two angular position sensors are faulty.

[0013] Preferably, the specific method for data modeling using convolutional neural networks and long short-term memory neural networks based on the data recorded by the angular position sensor in step S1 includes the following steps:

[0014] S11 Input Data Processing: The input data is a two-dimensional matrix, where each row represents a time window and each column contains multi-dimensional sensor data and environmental parameters;

[0015] S12 convolutional neural network feature extraction: The input data is convolved using 64 convolutional kernels, then an activation function is applied, followed by max pooling to reduce the size of the feature map, and a dropout function of 0.3 is added to prevent overfitting;

[0016] S13 Long Short-Term Memory Neural Network: A long short-term memory neural network is built on the output of a convolutional neural network for time series prediction. The long short-term memory neural network has 64 hidden neurons and finally uses a fully connected layer to output a value.

[0017] S14 Model Training: The Adam optimization algorithm is used to train the model on the data for 50 epochs, with each batch size of 100. The model records the loss values ​​on the training and test sets.

[0018] S15 model fusion: A pre-trained convolutional neural network is fused with an untrained long short-term memory neural network. The output of the convolutional neural network is used as the input of the long short-term memory neural network to achieve time series prediction.

[0019] S16 Data Denormalization: The model is trained on normalized data, and the prediction results need to be denormalized to obtain the original scale prediction values.

[0020] S17 Model Evaluation: The root mean square error is used as the primary evaluation metric to measure model performance, including training set error, validation set error, and prediction set error.

[0021] The beneficial effects of this invention are as follows: By using the method of this invention to detect the angular position sensor, when the angular position sensor fails, it is determined whether a single sensor or both sensors fail based on the threshold set by the convolutional neural network and the long short-term memory neural network. This allows for timely replacement of the angular position sensor, avoids data errors, and ensures the final detection result. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention;

[0023] Figure 2 A training flowchart is constructed for convolutional neural networks and long short-term memory neural networks. Detailed Implementation

[0024] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0025] The technical solution of this invention is as follows: a detection method for an angular position sensor, such as... Figure 1 As shown, it includes the following steps:

[0026] S1: Data modeling is performed using convolutional neural networks and long short-term memory neural networks based on the data recorded by the corner position sensors, and two corner position sensors are set up for detection;

[0027] S2: Based on the modeling data, detect whether the output values ​​of the two corner position sensors are maintained within the range and whether they are always fixed values ​​to determine whether the two corner position sensors are malfunctioning;

[0028] S3: Compare the predicted values ​​from the modeling data with the output values ​​of the two angle position sensors. If the output value of one sensor matches its corresponding predicted value, while the output value of the other sensor differs significantly from its corresponding predicted value, the system determines that one of the angle position sensors has failed. For example:

[0029] |S1-S 1_p |

[0030] |S2-S 2_p |≥i

[0031] At this point, it can be determined that sensor 2 has failed. Let S1 represent the signal value monitored by sensor 1, and S2 represent the signal value monitored by sensor 2. The data measured using a neural network is called S1. 1_p ,S 2_p ;

[0032] S4: When the output values ​​of both angular position sensors differ significantly from the predicted values, the error is calculated using the following formula:

[0033] ​

[0034]

[0035] In Formula 1, S1 represents the signal value of the first angle position sensor, and S2 represents the signal value of the second angle position sensor. denoted as the average value of the first and second angle position sensors, f represents the error caused by the angle position sensor malfunction, t represents time, and d represents the derivative function.

[0036] When the difference between the output values ​​of the two sensors and the predicted values ​​is large, and the output values ​​of the two sensors are also very different, it is no longer necessary to calculate the average value. Instead, we can study the changes in the f-values ​​of the two sensors separately.

[0037] S5: Determine whether the angular position sensor is faulty based on the change in the value of f. If the initial value of f jumps and the subsequent value of f returns to a stable state, i.e., σ(f) < γ, where γ is the error reference value, then the two angular position sensors are not faulty. Retrain the neural network based on the output value of the angular position sensor when the working conditions change drastically. After steps S3 and S4, if the value of f does not jump or the subsequent σ(f) > γ, then the two angular position sensors are faulty.

[0038] The specific method for data modeling using convolutional neural networks and long short-term memory neural networks based on the data recorded by the angular position sensor in step S1 is as follows: Figure 2 As shown, it includes the following steps:

[0039] S11 Input Data Processing: The input data is a two-dimensional matrix. The input data is the sensor recording data, which is divided into training set, validation set and test set according to the ratio of 60%, 20% and 20%. Each row represents a time window, and each column contains multi-dimensional sensor data and related environmental parameters, such as temperature, pressure and humidity.

[0040] S12 convolutional neural network feature extraction:

[0041] Define a network input sequence as:

[0042] X n×m =[x1,x2,...,x n-1 ,x n ]

[0043] Here, each x is an m-dimensional vector, resulting in an n-m two-dimensional matrix, where n is the number of values ​​per window and xi is the normalized value. The input data is convolved using 64 convolution kernels, followed by an activation function, max pooling to reduce the size of the feature map, and a dropout function of 0.3 to prevent overfitting.

[0044] Input features converted into a two-dimensional matrix are input into a convolutional neural network, j is the feature map index of each data window, using the values of the input data The output value of the l-th convolutional layer is calculated as:

[0045]

[0046] Wherein: represents the bias of the j-th feature map; W is the kernel weight; M is the filter size; σ is the activation function.

[0047] The pooling layer of the convolutional neural network reduces the spatial dimension of the representation to decrease the number of parameters and computational complexity of the network, and prevent overfitting. The calculation of the pooling layer is expressed as:

[0048]

[0049] Wherein: R is the pooling size, R<y; T is the step size for determining the pooling region, the convolutional neural network flattens all feature maps into one-dimensional vectors of the same number as the input of the long short-term memory neural network;

[0050] S13 Long Short-Term Memory Neural Network: a long short-term memory neural network is constructed on the basis of the output of the convolutional neural network for time series prediction, the long short-term memory neural network has 64 hidden neurons, and finally a fully connected layer is used to output one value;

[0051] S14 Model training: the Adam optimization algorithm is used to train for 50 epochs on the data, with each batch size being 100, and the model records loss values on the training set and the test set;

[0052] S15 Model fusion: the trained convolutional neural network is fused with the untrained long short-term memory neural network, and the output of the convolutional neural network is used as the input of the long short-term memory neural network to realize time series prediction;

[0053] The model has a time series prediction function, n groups of values before time t are used as input to predict the sensor output value after time t. Let x be the input, W be the weight matrix, h be the hidden layer information, b be the bias, the training process of LSTM is expressed by the following formula:

[0054] f t =σ(w f ·[h t-1, x t +b f ) (3)

[0055] i t =σ(wi •[h t-1 ,x t ]+b i (4)

[0056]

[0057] o t =σ(w o ·[h t-1 ,x t ]+b o (6)

[0058] h t =o t *tanh(c t (7)

[0059] In this context, equation (3) represents the "forgetting" operation of LSTM, equations (4) and (5) represent the "input" operation of the long short-term memory neural network, equation (6) represents the "update" operation of the long short-term memory neural network, and equations (6) and (7) represent the "output" operation of the long short-term memory neural network. The predicted value of the long short-term memory neural network is decoded by the fully connected layer to output the final result.

[0060] S16 Data Reverse Normalization: Because the model is trained on normalized data, the prediction results need to be reversed to obtain the original scale prediction values.

[0061] S17 Model Evaluation: The root mean square error is used as the main evaluation criterion to measure the model's performance, including training set error, validation set error, and prediction set error. The smaller the error value, the better the model fits.

[0062] This model combines convolutional neural networks and long short-term memory networks to process time series data and make predictions. It can be applied to predicting future values ​​of various time series sensor data.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A detection method for an angular position sensor, characterized in that, Includes the following steps: S1: Data modeling is performed using convolutional neural networks and long short-term memory neural networks based on the data recorded by the corner position sensors, and two corner position sensors are set up for detection; S2: Based on the modeling data, detect whether the output values ​​of the two corner position sensors are maintained within the range and whether they are always fixed values ​​to determine whether the two corner position sensors are malfunctioning; S3: Compare the predicted value of the modeling data with the output values ​​of the two corner position sensors. If the output value of one sensor matches its corresponding predicted value, while the output value of the other sensor differs significantly from its corresponding predicted value, the system determines that one of the corner position sensors has failed. S4: When the output values ​​of both angular position sensors differ significantly from the predicted values, the error is calculated using the following formula: in the formula The value represents the average of the first and second angle position sensors, f represents the error caused by the angle position sensor malfunction, and t represents time. S5: Determine whether the angular position sensor is faulty based on the change in the value of f. If the initial value of f jumps and the subsequent value of f returns to a stable state, i.e., σ(f) < γ, where γ is the error reference value, then the two angular position sensors are not faulty. Retrain the neural network based on the output value of the angular position sensor when the working conditions change drastically. After steps S3 and S4, if the value of f does not jump, or if the subsequent σ(f) > γ, then the two angular position sensors are faulty.

2. The detection method of an angular position sensor according to claim 1, characterized in that: The specific method for data modeling using convolutional neural networks and long short-term memory neural networks based on the data recorded by the angular position sensor in step S1 is as follows: Includes the following steps: S11 Input Data Processing: The input data is a two-dimensional matrix, where each row represents a time window and each column contains multi-dimensional sensor data and environmental parameters; S12 convolutional neural network feature extraction: The input data is convolved using 64 convolutional kernels, then an activation function is applied, followed by max pooling to reduce the size of the feature map, and a dropout function of 0.3 is added to prevent overfitting; S13 Long Short-Term Memory Neural Network: A long short-term memory neural network is built on the output of a convolutional neural network for time series prediction. The long short-term memory neural network has 64 hidden neurons and finally uses a fully connected layer to output a value. S14 Model Training: The Adam optimization algorithm is used to train the model on the data for 50 epochs, with each batch size of 100. The model records the loss values ​​on the training and test sets. S15 model fusion: A pre-trained convolutional neural network is fused with an untrained long short-term memory neural network. The output of the convolutional neural network is used as the input of the long short-term memory neural network to achieve time series prediction. S16 Data Denormalization: The model is trained on normalized data, and the prediction results need to be denormalized to obtain the original scale prediction values. S17 Model Evaluation: The root mean square error is used as the primary evaluation metric to measure model performance, including training set error, validation set error, and prediction set error.

Images