Speed reducer sensor layout optimization method, system and equipment and storage medium

Abstract

The invention belongs to the technical field of fault testing, and provides a speed reducer sensor layout optimization method, system and device and a storage medium, and the method comprises the steps: carrying out the local linear processing of fault features, and obtaining a reduced fault feature matrix; measuring the sensitivity of the measuring points based on the fault feature matrix; and judging whether the fault sensitivity degree of the measurement point meets a target or not based on the measurement point sensitivity measurement, if so, ending, otherwise, resetting the measurement point until the fault sensitivity degree of the measurement point meets the target. A plurality of measuring points are arranged on a speed reducer fault simulation experiment platform, then fault features of the measuring points are extracted, local linear embedding processing is carried out on the fault features to obtain a reduced three-dimensional vector, the convolutional neural network is used for carrying out index evaluation on the fault sensitivity degree of the measuring points, and the fault sensitivity degree of the speed reducer is evaluated. Therefore, the sensor layout is optimized based on the evaluation result, and the accuracy and efficiency of fault monitoring are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault testing, and in particular relates to a speed reducer sensor layout optimization method, system, device and storage medium. Background Art

[0002] When monitoring helicopter main transmission systems, optimizing sensor layout is crucial for improving the accuracy and efficiency of fault detection. In other fields, such as industrial production equipment, environmental monitoring systems, and traffic monitoring systems, multiple sensors are deployed to collect data to monitor the system's operating status and performance. However, the layout of sensors directly affects the accuracy and reliability of system status monitoring. Traditional sensor layout methods are often based on experience, simulation, or uniform distribution principles, failing to fully consider the actual data fault characteristics at different locations in the system and their impact on monitoring results. This can lead to problems such as irrational sensor layout, wasted resources, or the omission of critical fault information.

[0003] A helicopter main reducer is a complex mechanical component that operates in harsh environments, subjecting it to high speeds, high torques, heavy loads, and complex forces. This makes it prone to various faults during operation, such as gear wear, bearing damage, and shaft deformation. Different fault types produce distinct physical signal signatures at different locations. By optimizing sensor layout, sensors can be placed in locations most sensitive to fault signatures, enabling them to more accurately capture signals reflecting the fault.

[0004] Existing technologies for optimizing helicopter main reducer sensor placement have several shortcomings. Traditional optimization algorithms, such as the Effective Independent (EFI) method and the Kinetic Energy Method (KEM), each have their own limitations. For example, they may generate information redundancy in local areas, hindering damage localization. These methods may not fully capture key fault characteristics, resulting in incomplete monitoring data and inaccurate diagnosis. Sensor placement optimization based on the traditional Fisher Information Criterion can lead to localized clustering of measurement points, resulting in information redundancy and hindering damage localization. This means that sensors are overly concentrated in some areas, while others may lack adequate monitoring, resulting in imbalanced monitoring data. Most existing sensor placement optimization methods for structural damage identification only consider a single optimization objective and fail to balance the multiple conflicts between the sensitivity and independence of multiple sensors to local damage. This limits the multi-objective and comprehensive nature of sensor placement optimization. These issues restrict the effectiveness and scope of sensor placement optimization, leading to incomplete monitoring data and inaccurate diagnosis, increasing costs and maintenance difficulties, and preventing the full realization of the optimal performance of a structural health monitoring system composed of multiple sensors. Summary of the Invention

[0005] To solve the problems in the background art, the present invention proposes an optimization method, system, device and storage medium for the layout of reducer sensors.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An optimization method for the layout of reducer sensors includes the following steps:

[0008] Set a number of measurement points on the reducer fault simulation experiment platform;

[0009] Collect the vibration signals of all measurement points;

[0010] Calculate the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain the fault characteristics;

[0011] Perform local linear processing based on the fault characteristics to obtain a reduced fault characteristic matrix;

[0012] Measure the sensitivity of the measurement points based on the fault characteristic matrix;

[0013] Judge whether the fault sensitivity of the measurement points meets the target based on the measurement of the sensitivity of the measurement points. If it meets the target, end. Otherwise, reset the measurement points until the fault sensitivity of the measurement points meets the target.

[0014] Preferably, collecting the vibration signals through the reducer fault simulation experiment platform includes the following steps:

[0015] Simulate the speed-up and speed-down processes and steady-state operating conditions of the helicopter main reduction planetary gear train of the reducer fault simulation experiment platform;

[0016] Collect the vibration signals of the measurement points on the reducer fault simulation experiment platform during the simulation process.

[0017] Preferably, calculating the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain the fault characteristics includes the following steps:

[0018] Calculate the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis and skewness in the time-domain characteristic indexes based on the vibration signals;

[0019] Calculate the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis and spectral skewness in the frequency-domain characteristic indexes based on the vibration signals;

[0020] Extract the fault characteristics in the time-domain characteristic indexes and frequency-domain characteristic indexes based on the calculation results.

[0021] Preferably, local linear processing is performed based on the fault features to obtain a reduced fault feature matrix, including the following steps:

[0022] Set the data set of fault features;

[0023] Construct a local reconstruction weight matrix based on the data set of fault features and minimize the loss function for the first time;

[0024] After the first minimization of the loss function, calculate the best reconstructed low-dimensional embedding vector and minimize the loss function for the second time.

[0025] Preferably, the first minimization of the loss function includes:

[0026]

[0027] In the formula, W = ω ij = [ω1, ω2, …, ω N is the local reconstruction weight matrix; g ij is the k nearest neighbor points of g i ; g i is the sample of the fault feature; N represents the number of samples, N ≥ n ≥ 1; minε(W) represents minimizing the loss function ε(W) with respect to W; I cbd is the identity matrix; T represents the transpose matrix;

[0028] The second minimization of the loss function includes:

[0029]

[0030] Y is the mapping vector of the data set of fault features in the low-dimensional space; y ij is the k nearest neighbor points of y i ; y i is the mapping of the sample of the fault feature in the low-dimensional space; minψ(Y) represents minimizing the loss function ψ(W) with respect to Y.

[0031] Preferably, the measurement point sensitivity metric is performed based on the fault feature matrix, including the following steps:

[0032] Divide the fault feature matrix into a training set, a validation set, and a test set;

[0033] Perform data augmentation on the training set and the validation set;

[0034] Use the augmented training set to train the convolutional neural network;

[0035] Use the augmented validation set to validate the trained convolutional neural network;

[0036] Test the verified convolutional neural network using the test set;

[0037] Apply the Gradient-Weighted Class Activation Mapping method to the samples in the test set to generate activation maps of different convolutional layers for different fault features;

[0038] Analyze the activation maps and perform measurement point sensitivity metrics.

[0039] A reducer sensor layout optimization system, comprising:

[0040] A setting unit for setting a number of measurement points on the reducer fault simulation experiment platform;

[0041] An acquisition unit for acquiring vibration signals of all measurement points;

[0042] A calculation unit for calculating time-domain feature indicators and frequency-domain feature indicators based on the vibration signals to obtain fault features;

[0043] A reduction unit for performing local linear processing based on the fault features to obtain a reduced fault feature matrix;

[0044] A metric unit for performing measurement point sensitivity metrics based on the fault feature matrix;

[0045] A feedback unit for judging whether the fault sensitivity of the measurement points meets the target based on the measurement point sensitivity metrics. If the target is met, the process ends; otherwise, the measurement points are reset until the fault sensitivity of the measurement points meets the target.

[0046] Preferably, the acquisition unit includes:

[0047] A simulation module for simulating the speed-up and speed-down processes and steady-state operating conditions of the helicopter main reduction planetary gear system of the reducer fault simulation experiment platform;

[0048] An acquisition module for acquiring vibration signals during the simulation process of the reducer fault simulation experiment platform.

[0049] Preferably, the calculation unit includes:

[0050] A first calculation module for calculating the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis, and skewness in the time-domain feature indicators based on the vibration signals;

[0051] A second calculation module for calculating the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis, and spectral skewness in the frequency-domain feature indicators based on the vibration signals;

[0052] An extraction module for extracting fault features in time-domain feature indicators and frequency-domain feature indicators based on calculation results.

[0053] Preferably, the reduction element includes:

[0054] A setting module for setting a data set of fault features;

[0055] A first reduction module for constructing a local reconstruction weight matrix based on the data set of fault features and minimizing the loss function for the first time;

[0056] A second reduction module for calculating the low-dimensional embedding vector of the best reconstruction and minimizing the loss function for the second time after the first minimization of the loss function.

[0057] A device includes a memory and a processor;

[0058] The memory is used to store a computer program;

[0059] When the processor is used to execute the program stored on the memory, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0060] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0061] A computer program product includes a computer program / instructions, and when the computer program / instructions are executed by a processor, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0062] Advantages of the present invention:

[0063] 1. In the present invention, multiple measurement points are set on a reducer fault simulation experimental platform, and then the fault features of each measurement point are extracted. Through local linear embedding processing of the fault features, a reduced three-dimensional vector is obtained, and a convolutional neural network is used to evaluate the index of the sensitivity of the measurement points to faults, so as to optimize the sensor layout based on the evaluation results, improving the accuracy and efficiency of fault monitoring.

[0064] 2. The optimization method of the present invention considers multiple optimization objectives, can balance various contradictions such as the sensitivity and independence of multiple sensors to local damage, is beneficial to the multi-objectivity and comprehensiveness of sensor layout optimization, expands the effect and application range of sensor layout optimization, makes the monitoring data more complete and improves the diagnostic accuracy, reduces the cost and maintenance difficulty, and fully exerts the optimal efficiency of the structural health monitoring system composed of multiple sensors.

[0065] Other features and advantages of the present invention will be set forth in the following description, and in part will be obvious from the description, or may be learned by practice of the present invention. The objectives and other advantages of the present invention may be realized and attained by the structure particularly pointed out in the specification and the drawings. Description of the Drawings

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following briefly introduces the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings may be obtained based on these drawings.

[0067] Figure 1 The flowchart showing an optimization method for the layout of a reducer sensor according to the present invention is shown;

[0068] Figure 2 The flowchart showing the construction of the fusion health index for a helicopter transmission system according to the present invention is shown;

[0069] Figure 3 The structural diagram of the convolutional neural network according to the present invention is shown;

[0070] Figure 4 The fault signal diagram of the helicopter transmission system obtained by actual acquisition according to the present invention is shown, where (a) is the time-domain signal diagram, (b) is the frequency-domain signal diagram, and (c) is the envelope spectrum signal diagram;

[0071] Figure 5 The health index diagram calculated according to the present invention is shown;

[0072] Figure 6 The structural diagram of an optimization system for the layout of a reducer sensor according to the present invention is shown. Detailed Embodiments

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following clearly and completely describes the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0074] As Figure 1 shown, an optimization method for the layout of a reducer sensor includes the following steps:

[0075] S1: Set a number of measurement points on the reducer fault simulation experiment platform, and then collect the vibration signals of all measurement points.

[0076] S2: Calculate the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signal to obtain the fault characteristics.

[0077] S3: Perform local linear processing based on the fault characteristics to obtain the reduced fault characteristic matrix.

[0078] S4: Measure the sensitivity of the measuring points based on the fault characteristic matrix.

[0079] S5: Judge whether the fault sensitivity of the measuring point meets the target based on the measurement point sensitivity measurement. If the target is met, end. Otherwise, reset the measuring point until the fault sensitivity of the measuring point meets the target.

[0080] It should be noted that S1-S5 simulate the fault conditions that the reducer may encounter during actual operation, so as to collect the vibration signals containing fault information, providing key data for subsequent feature extraction and diagnosis. Subsequently, the reduced fault characteristic matrix is obtained by processing the vibration information. This matrix is more concise and efficient, which is beneficial to subsequent diagnosis and sensor layout optimization. Finally, the measuring points with high sensitivity can more accurately reflect the operating state and fault conditions of the reducer, and these measuring points can be given priority in the sensor layout optimization. This method actually uses machine deep learning to optimize the sensor layout, which can improve the accuracy and efficiency of fault diagnosis and reduce the maintenance cost.

[0081] Next, taking the sensor layout optimization of the helicopter main reducer as an example, S1-S5 will be further described.

[0082] As an optional solution, S1 includes the following steps:

[0083] S101: Design a reducer fault simulation test platform, which includes a drive system, a transmission system, a loading system, a support system, a measurement and control system, and a gear lubrication system.

[0084] S102: Set a number of measurement points on the surface of the reducer fault simulation experiment platform, generally not less than 8.

[0085] S103: Simulate the speed-up and speed-down process and steady-state operating conditions of the planetary gear train of the helicopter main reduction gear of the reducer fault simulation experiment platform.

[0086] S104: Collect the vibration signals of the reducer fault simulation experiment platform during the simulation process.

[0087] It should be noted that in S1, a reduction gearbox fault simulation test platform is designed. By setting multiple measurement points, simulating the process of speed increase and decrease and the steady-state operation condition, and collecting vibration signals, the operation condition of the planetary gear train of the helicopter main reduction gearbox can be comprehensively and accurately simulated, which helps to timely detect potential faults and improve the accuracy and efficiency of fault diagnosis. At the same time, the test platform also has a gear lubrication system, which can simulate the gear operation state under different lubrication conditions and provide data support for optimizing gear lubrication and reducing wear.

[0088] As an alternative solution, S2 includes the following steps:

[0089] S201: Calculate the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis, and skewness in the time-domain characteristic indexes based on the vibration signal.

[0090] S202: Calculate the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis, and spectral skewness in the frequency-domain characteristic indexes based on the vibration signal.

[0091] S203: Extract the fault characteristics in the time-domain characteristic indexes and frequency-domain characteristic indexes based on the calculation results.

[0092] It should be noted that X = [x(1),..., x(n),..., x(N)], X can represent the set of acceleration signals (vibration signals) collected, where N is the sampling length of the signal, that is, the number of samples. The formula for calculating the average energy in S201 is as follows:

[0093]

[0094] The formula for calculating the standard deviation is as follows:

[0095]

[0096] The formula for calculating the root mean square amplitude is as follows:

[0097]

[0098] The formula for calculating the root mean square value x rms is as follows:

[0099]

[0100] The formula for calculating the peak value x p is as follows:

[0101] x p = max|x(n)|; (5)

[0102] The formula for calculating the peak index C is as follows:

[0103]

[0104] The formula for calculating the margin index L is as follows:

[0105]

[0106] The formula for calculating kurtosis is as follows:

[0107]

[0108] The formula for calculating skewness is as follows:

[0109]

[0110] In S202, perform spectral analysis on the collected acceleration signal to obtain the frequency-domain data S = [s(1),..., s(k),..., s(K)], where the formula for calculating the spectral mean F1 is as follows:

[0111]

[0112] The formula for calculating the frequency centroid F2 is as follows:

[0113]

[0114] The formula for calculating the root mean square frequency F3 is as follows:

[0115]

[0116] The formula for calculating the standard deviation frequency F4 is as follows:

[0117]

[0118] The formula for calculating the peak frequency F5 is as follows:

[0119]

[0120] where f k represents the frequency value corresponding to the k-th frequency-domain data. The formula for calculating the amplitude spectrum F6 is as follows:

[0121]

[0122] The formula for calculating the phase spectrum F7 is as follows:

[0123]

[0124] The formula for calculating the power spectral density F8 is as follows:

[0125]

[0126] The formula for calculating the resonance peak F9 is as follows:

[0127]

[0128] The formula for calculating the spectral entropy F10 is as follows:

[0129]

[0130] The formula for calculating the spectral kurtosis F11 is as follows:

[0131]

[0132] The formula for calculating the spectral skewness F12 is as follows:

[0133]

[0134] As an alternative, S3 includes the following steps:

[0135] S301: Set the dataset G of fault features, which can be set as:

[0136]

[0137] where R D represents all the sample data.

[0138] Search for the k nearest neighbor points of each fault feature sample point g i in the high-dimensional space according to the Euclidean distance:

[0139] N i = KNN(g i , k) = [g ij , L, g ik ; (23)

[0140] Then the sample point can be represented by a linear combination of the k nearest neighbor data points; where N i represents the data set of the k nearest neighbor points of the fault feature sample point g i ; KNN represents the K-nearest neighbor algorithm.

[0141] S302: Calculate the reconstruction weights in the sample point neighborhood, mainly by constructing a local reconstruction weight matrix based on the dataset of fault features and minimizing the loss function for the first time, including:

[0142]

[0143] where W = ω ij = [ω1, ω2,..., ω N is the local reconstruction weight matrix; gij are the k nearest neighbor points of g i ; where g i is a sample of the fault feature; N represents the number of samples, N≥n≥1; minε(W) represents minimizing the loss function ε(W) with respect to W; I cbd is the identity matrix, used to constrain the calculation result of

[0144] S303: After the first minimization of the loss function, calculate the best reconstructed low-dimensional embedding vector. Since the reconstruction weights describe the inherent geometric properties of each neighborhood and do not depend on a specific reference frame. So the reconstruction weight matrix calculated by (24) is still valid. Minimize the loss function again.

[0145] Where the second minimization of the loss function includes:

[0146]

[0147] Y is the mapping vector of the dataset of the fault feature in the low-dimensional space; y ij are the k nearest neighbor points of y i ; y i is the mapping of the sample of the fault feature in the low-dimensional space; minψ(Y) represents minimizing the loss function with respect to Y.

[0148] It should be noted that, first, in S301, the k nearest neighbor points of each sample point are searched in the high-dimensional space according to the Euclidean distance, realizing the accurate capture and description of the fault feature. This method can make full use of the information in the dataset and improve the accuracy of fault recognition. Second, by calculating the reconstruction weights of the sample point neighborhood, constructing the local reconstruction weight matrix, and minimizing the loss function for the first time, the local geometric structure of the dataset can be effectively retained. This helps to maintain the intrinsic characteristics of the data in the subsequent dimensionality reduction process and improve the dimensionality reduction effect. Finally, by minimizing the loss function again and calculating the best reconstructed low-dimensional embedding vector, the dimensionality reduction result can be further optimized. Since the reconstruction weights describe the inherent geometric properties of each neighborhood and do not depend on a specific reference frame, this method has good robustness and applicability. The S3 scheme realizes the effective dimensionality reduction of high-dimensional data and fault recognition by accurately capturing the fault feature, retaining the local geometric structure of the dataset, and optimizing the dimensionality reduction result, providing strong support for subsequent fault diagnosis and prediction.

[0149] As an optional scheme, S4 includes the following steps:

[0150] S401: Divide the fault feature matrix into a training set, a validation set, and a test set, which can be divided according to the ratio of 70%:15%:15%, while ensuring that each set contains data under different fault conditions to ensure the generalization ability of the model.

[0151] S402: Perform data augmentation on the training set and the validation set, such as adding noise, time translation, time scaling, etc., to increase the diversity of the data and improve the robustness of the model.

[0152] S403: Use the augmented training set to train the convolutional neural network.

[0153] S404: Validate the trained convolutional neural network using the augmented validation set.

[0154] S405: Test the validated convolutional neural network using the test set, for example, calculate the loss and accuracy on the test set to obtain the performance of the model on unknown data.

[0155] S406: Apply the Gradient-weighted Class Activation Mapping (Grad-CAM) method to the samples in the test set to generate activation maps of different convolutional layers for different fault features, and analyze the sensitivity of different convolutional layers to different fault classes.

[0156] S407: Analyze the activation maps and perform measurement point sensitivity metrics.

[0157] It should be noted that S4 improves the performance and generalization ability of the fault diagnosis model through a systematic process. By scientifically dividing the fault feature matrix into a training set, a validation set, and a test set, and ensuring that each set contains data under different fault conditions, the generalization ability of the model is effectively enhanced. The data augmentation step significantly increases the diversity of the data and further improves the robustness of the model. Using the augmented data for training and validation ensures the accuracy of the model under complex fault conditions. Finally, by applying the Grad-CAM method to generate activation maps and performing measurement point sensitivity metrics, the sensitivity of the model to different fault features is deeply analyzed, providing a strong basis for fault diagnosis. Among them, the data augmentation step is particularly helpful for improving the performance of the model under diverse and complex fault conditions. In S4, the convolutional neural network is adjusted according to actual needs, and different feature relationships are evaluated through the gradient-weighted class mapping method, which associates its classification decision with the input space position of the neural network, thereby realizing sensitivity metrics.

[0158] As an optional solution, S5 includes the following steps:

[0159] S501: Determine whether the measurement point sensitivity metric is within the target sensitivity range, and this target sensitivity range can be set according to the operating conditions of the device.

[0160] S502: If it is within the target sensitivity range, it means that the position of the measurement point can represent the sensitivity of the device, and the whole process ends. If it is not within the target sensitivity range, it means that the position of the measurement point cannot represent the sensitivity of the device, so it is necessary to return to step S1 and reset the measurement point. Figure 2 ,After processing by the Grad-CAM method, an evaluation result of the sensor fault ,sensitivity will be generated, which will be fed back to the initial step.

[0161] like Figure 3 Figure 1 shows the convolutional neural network architecture of the present invention, in which the original signals (such as the training set, validation set, and test set) are sequentially passed through four convolutional basic modules and a fully connected module to obtain the classification results. Each convolutional basic module includes a one-dimensional convolution layer, a batch normalization unit, a rectified linear unit, and a maximum pooling layer. The fully connected module includes a tiling layer, a random dropout unit, and a fully connected layer.

[0162] The following is a specific implementation based on the process of S1-S4.

[0163] Step 1: Set up measurement points and collect data. The helicopter transmission system is a critical component, and its operating status directly impacts flight safety. Optimizing sensor layout is crucial for effective transmission system monitoring and fault diagnosis, which relies on comprehensive and accurate data collection. For core components such as the main reducer, tail reducer, and drive shaft, identify areas prone to failure, such as the main reducer's gear meshing and bearing mounting areas, the tail reducer's output shaft, and the drive shaft's connection points. These areas should be the focus of data collection. Select a high-precision accelerometer with a frequency response range that meets the helicopter transmission system's operating frequency requirements. A range of 0-10kHz or wider is generally recommended, with measurement accuracy exceeding ±0.1m / s². This allows for accurate capture of minute vibration changes. Piezoelectric accelerometers, for example, offer high sensitivity and excellent dynamic response. Select a data acquisition instrument with multi-channel simultaneous acquisition capabilities. The number of channels should meet the sensor requirements, with at least 16 channels and a sampling frequency of at least 20kHz. This ensures simultaneous acquisition of data from different sensors without losing high-frequency information. At the same time, the data acquisition instrument should have good anti-interference capabilities, and the built-in filter can effectively remove external electromagnetic interference to ensure the purity of the collected data. Such technology is not the core innovation of this invention, so it will not be described in detail.

[0164] Step 2: Calculate the time domain signal characteristic index and the signal frequency domain characteristic index, see formula (1)-(21) for details.

[0165] Step 3: Perform local linear processing on the feature quantity.

[0166] Step 4: Split the training set, validation set, and test set in the ratio of 70%:15%:15% according to the features. The neural network parameters are set as follows in the table:

[0167] Table 1

[0168]

[0169] Note: "~" indicates that the parameter does not exist in the max pooling layer.

[0170] It should be noted that before updating the diagnostic model parameters, the weights and biases of each node are initialized using Kaiming (He Kaiming) initialization. During the update process of the diagnostic model parameters, the number of model parameter updates is set to 400 times, and the batch processing number during each update process is set to 20; the adaptive moment estimation (adam, Adaptive Moment Estimation) optimization algorithm is used to update the network weights and biases to minimize the constructed integrated loss function. In the Adam optimization algorithm, the initial learning rate is set to 0.001, and its value is reduced to 0.0001 after 200 times of model parameter updates. Since the input data dimension of the diagnostic model is relatively large, the recognition performance tests in the experimental cases are all completed on a workstation equipped with a Graphics Processing Unit (GPU). To reduce the influence of the random initialization of the diagnostic model parameters on the diagnostic results, each group of comparative experiments in the subsequent experimental cases is executed ten times. Subsequently, gradient-weighted class activation mapping is carried out to search for features, and different sensor fault sensitive feature parameters are obtained, so as to provide guidance for optimizing the layout.

[0171] Step 5: Based on the results of Step 4, determine whether the sensitivity of the measurement points meets the requirements, and select whether to reset the measurement points according to the judgment results.

[0172] As Figure 4 shown, it is the fault signal of the helicopter transmission system obtained by real acquisition, where Figure 4 (a) is the relationship between time and acceleration of the time-domain signal, Figure 4 (b) is the relationship between the meshing frequency order and acceleration of the frequency-domain signal, Figure 4 (c) is the relationship between the frequency and acceleration of the envelope spectrum signal.

[0173] As Figure 5 shown, it is the calculated health index of each measurement point. Among them, the correlation between the frequency and amplitude of the four measurement points is related to the position of the measurement points on the experimental platform, and the amplitude changes with the frequency following a certain pattern.

[0174] As Figure 6As shown in the figure, it is a system for optimizing the layout of a reducer sensor, including a setting unit, a collection unit, a calculation unit, a reduction unit, a measurement unit, and a feedback unit. Among them, the setting unit is used to set a number of measurement points on the reducer fault simulation experiment platform; the collection unit is used to collect the vibration signals of all measurement points; the calculation unit is used to calculate the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain fault characteristics; the reduction unit is used to perform local linear processing based on the fault characteristics to obtain a reduced fault characteristic matrix; the measurement unit is used to measure the sensitivity of the measurement points based on the fault characteristic matrix; the feedback unit judges whether the fault sensitivity of the measurement points meets the target based on the measurement of the measurement point sensitivity. If the target is met, the process ends. Otherwise, the measurement points are reset until the fault sensitivity of the measurement points meets the target.

[0175] As a preferred solution, the collection unit includes a simulation module and a collection module. Among them, the simulation module is used to simulate the speed-up and speed-down process and the steady-state operation condition of the helicopter main reduction planetary gear system of the reducer fault simulation experiment platform; the collection module is used to collect the vibration signals during the simulation process of the reducer fault simulation experiment platform.

[0176] As a preferred solution, the calculation unit includes a first calculation module, a second calculation module, and an extraction module. Among them, the first calculation module is used to calculate the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis, and skewness in the time-domain characteristic indexes based on the vibration signals; the second calculation module is used to calculate the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis, and spectral skewness in the frequency-domain characteristic indexes based on the vibration signals; the extraction module is used to extract the fault characteristics in the time-domain characteristic indexes and frequency-domain characteristic indexes based on the calculation results.

[0177] As a preferred solution, the reduction unit includes a setting module, a first reduction module, and a second reduction module. Among them, the setting module is used to set the data set of the fault characteristics; the first reduction module is used to construct a local reconstruction weight matrix based on the data set of the fault characteristics and minimize the loss function for the first time; the second reduction module is used to calculate the optimal reconstructed low-dimensional embedding vector after the first minimization of the loss function and minimize the loss function for the second time.

[0178] A device includes a memory and a processor;

[0179] The memory is used to store computer programs;

[0180] When the processor is used to execute the program stored on the memory, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0181] It should be noted that the memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory.

[0182] The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0183] A computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0184] It should be noted that the computer-readable storage medium may be included in the device / apparatus described in the above embodiments; it may also exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs. When the one or more programs are executed, it implements a method for optimizing the layout of a reducer sensor according to an embodiment of the present invention.

[0185] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, device, or apparatus.

[0186] A computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, it implements the above-mentioned method for optimizing the layout of a reducer sensor.

[0187] Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the layout of a reducer sensor, characterized in that, It includes the following steps: Set several measurement points on the reducer fault simulation experimental platform; Collect the vibration signals of all measurement points; Calculate the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain the fault characteristics; Perform local linear processing based on the fault characteristics to obtain the reduced fault characteristic matrix; Perform measurement point sensitivity measurement based on the fault characteristic matrix; Judge whether the fault sensitivity of the measurement points meets the target based on the measurement point sensitivity measurement. If the target is met, end. Otherwise, reset the measurement points until the fault sensitivity of the measurement points meets the target.

2. The method for optimizing the layout of a reducer sensor according to claim 1, wherein Collect the vibration signals of all measurement points, including the following steps: Simulate the speed-up and speed-down processes and steady-state operating conditions of the helicopter main reduction planetary gear train on the reducer fault simulation experimental platform; Collect the vibration signals of the measurement points on the reducer fault simulation experimental platform during the simulation process.

3. The method for optimizing the layout of a reducer sensor according to claim 1, characterized in that Calculate the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain the fault characteristics, including the following steps: Calculate the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis, and skewness in the time-domain characteristic indexes based on the vibration signals; Calculate the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis, and spectral skewness in the frequency-domain characteristic indexes based on the vibration signals; Extract the fault characteristics in the time-domain characteristic indexes and frequency-domain characteristic indexes based on the calculation results.

4. A method for optimizing the layout of a speed reducer sensor according to claim 1, characterized in that, Perform local linear processing based on the fault characteristics to obtain the reduced fault characteristic matrix, including the following steps: Set the data set of the fault characteristics; Construct a local reconstruction weight matrix based on the data set of the fault characteristics and minimize the loss function for the first time; After the first minimization of the loss function, calculate the optimal reconstructed low-dimensional embedding vector and minimize the loss function for the second time.

5. The method for optimizing the layout of a reducer sensor according to claim 4, characterized in that The first minimization of the loss function includes: Where, W = ω ij = [ω1, ω2, …, ω N is the local reconstruction weight matrix; g ij is the k nearest neighbor points of g i ; g i is the sample of the fault feature; N represents the number of samples, N ≥ n ≥ 1; minε(W) represents minimizing the loss function ε(W) with respect to W; I cbd is the identity matrix; T represents the transpose matrix; The second minimization of the loss function includes: Y is the mapping vector of the dataset of fault features in the low-dimensional space; y ij is y i 's k nearest neighbor points; y i The mapping of the samples of fault features in the low-dimensional space; minψ(Y) represents minimizing the loss function with respect to Y.

6. A method for optimizing the layout of a reducer sensor according to claim 1, characterized in that Perform measurement point sensitivity measurement based on the fault characteristic matrix, including the following steps: Divide the fault characteristic matrix into a training set, a validation set, and a test set; Perform data augmentation on the training set and the validation set; Train the convolutional neural network using the data-augmented training set; Validate the trained convolutional neural network using the data-augmented validation set; Test the validated convolutional neural network using the test set; Apply the gradient-weighted class activation mapping method to the samples in the test set to generate the activation maps of different convolutional layers for different fault characteristics; Analyze the activation maps and perform measurement point sensitivity measurement.

7. A reducer sensor layout optimization system, characterized in that, It includes: A setting unit for setting several measurement points on the reducer fault simulation experimental platform; A collection unit for collecting the vibration signals of all measurement points; A calculation unit for calculating the time-domain characteristic indexes and frequency-domain characteristic indexes based on the vibration signals to obtain the fault characteristics; A reduction unit for performing local linear processing based on the fault characteristics to obtain the reduced fault characteristic matrix; A measurement unit for performing measurement point sensitivity measurement based on the fault characteristic matrix; A feedback unit determines whether the fault sensitivity of a measurement point meets a target based on the measurement point sensitivity metric. If the target is met, the process ends; otherwise, the measurement point is reset until the fault sensitivity of the measurement point meets the target.

8. The optimized system for the layout of a speed reducer sensor according to claim 7, wherein The acquisition unit includes: An analog module for simulating the speed-up and speed-down processes and steady-state operating conditions of the helicopter main reduction planetary gear train of the reducer fault simulation experiment platform; An acquisition module for acquiring vibration signals during the simulation process of the reducer fault simulation experiment platform.

9. The optimized system for the layout of a speed reducer sensor according to claim 7, characterized in that The calculation unit includes: A first calculation module for calculating the average energy, standard deviation, root mean square amplitude, root mean square value, peak value, peak index, margin index, kurtosis, and skewness in the time-domain characteristic indicators based on the vibration signal; A second calculation module for calculating the spectral mean, spectral centroid, root mean square frequency, standard deviation frequency, peak frequency, amplitude spectrum, phase spectrum, power spectral density, resonance peak, spectral entropy, spectral kurtosis, and spectral skewness in the frequency-domain characteristic indicators based on the vibration signal; An extraction module for extracting fault features from the time-domain characteristic indicators and frequency-domain characteristic indicators based on the calculation results.

10. The optimized system for the layout of a reducer sensor according to claim 7, characterized in that, The reduction element includes: A setting module for setting a dataset of fault features; A first reduction module for constructing a local reconstruction weight matrix based on the dataset of fault features and minimizing the loss function for the first time; A second reduction module for calculating the optimal reconstructed low-dimensional embedding vector after the first minimization of the loss function and minimizing the loss function for the second time.

11. A device, characterized in that, It includes a memory and a processor; The memory is used to store a computer program; When the processor executes the program stored in the memory, it implements a method for optimizing the layout of sensors for a reducer according to any one of claims 1-6.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for optimizing the layout of sensors for a reducer according to any one of claims 1-6.

13. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, it implements a method for optimizing the layout of sensors for a reducer according to any one of claims 1-6.

Images