An automatic noise-reduction vibration signal acquisition system and method

Abstract

The automatic noise reduction vibration signal acquisition system comprises a sensor module, a clock chip, an analog-to-digital converter, a programmable logic device, a digital signal processing chip, a neural network processor and a JTAG interface. The sensor module is used for collecting vibration signals and inputting the vibration signals in the form of analog signals into the analog-to-digital converter. The clock chip is used for providing a running clock signal for the system. The programmable logic device is electrically connected with the analog-to-digital converter and provides a sampling clock signal for the analog-to-digital converter. After receiving the sampling clock signal, the analog-to-digital converter converts the analog signal at the sensor end into a digital signal and transmits the digital signal into the digital signal processing chip. The neural network processor writes the network structure of a blind noise self-encoder. The JTAG interface is used for programming and debugging of each component in the system. The automatic noise reduction acquisition system has a simple noise reduction process and does not need signal prior information or manually set noise reduction parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of signal acquisition and noise reduction, and in particular to a vibration signal acquisition system and method with automatic noise reduction. Background Art

[0002] Vibration signals are an important carrier of structural state information. Condition monitoring and fault diagnosis in construction and mechanical engineering rely heavily on vibration signal analysis. However, the noise generated by sensors during vibration signal acquisition can significantly reduce the accuracy and precision of signal analysis results. To reduce the impact of noise on signal analysis, the original signal must first be subjected to noise reduction processing. Common signal noise reduction methods typically require knowledge of certain prior information or the manual setting of noise reduction parameters. For example, frequency domain filtering methods require setting the frequency domain range for noise reduction based on the frequency domain characteristics of the signal; wavelet noise reduction methods require the selection of appropriate wavelet packets and noise reduction thresholds. Therefore, developing a vibration signal acquisition system with automatic noise reduction capabilities to achieve blind noise reduction of vibration signals without prior information can better meet the needs of collecting vibration signals from various practical engineering structures. Summary of the Invention

[0003] In view of the problems in the background, an object of the present invention is to provide a vibration signal acquisition system with automatic noise reduction, the vibration signal acquisition system with automatic noise reduction comprising:

[0004] The main components of the vibration signal acquisition system with automatic noise reduction proposed in the present invention include:

[0005] The sensor module is used to collect vibration signals and connect the vibration signals to the digital-to-analog converter module in the form of analog signals.

[0006] The clock chip is mainly composed of components such as crystal oscillator circuit and charging circuit, which provides the operating clock signal for the system.

[0007] The analog-to-digital converter (ADC) receives the sampling clock signal and converts the analog signal from the sensor into a digital signal and transmits it to the digital signal processing chip. In the pre-collection phase of the system, the double sampling of the present invention is performed in conjunction with the above.

[0008] Programmable logic device (CPLD), the main components of which are control circuit, counter and pulse synthesis circuit, etc., provide sampling clock signal (uniform sampling or double sampling) for analog-to-digital converter. The generation process of sampling clock signal is as follows: the clock circuit generates the sampling clock signal according to the set sampling frequency. Generate uniform sampling clock signals and send them to counter 1 and counter 2 respectively. After the counter is started, its high-frequency clock continues to count and outputs a pulse signal when the counter overflows. Figure 4 As shown, when double sampling is required, the control circuit will open the delay switch to generate a delay (Δt<1 / fs ) sequence is sent to counter 2, causing the pulse signal output by counter 2 to be delayed by a time interval Δt compared to counter 1. The pulses generated by the two counters form the required double-sampled pulse signal after passing through the pulse synthesis circuit. In addition, when double sampling is not required, the control circuit does not output a delay signal. At this time, the pulses generated by the two counters overlap, and the clock signal after passing through the pulse synthesis circuit has a frequency of f s The uniform sampling signal, uniform sampling and double sampling effects are as follows Figure 5 shown.

[0009] The digital signal processing chip (DSP) is a microprocessor used for system control and data processing. Its special functions in the present invention are: (1) controlling the sampling system. The control clock module generates a sampling clock signal according to the set sampling frequency and inputs it into the CPDL; the double sampling is controlled by controlling the switch of the CPLD counter circuit delay. (2) data processing. The double sampling data is processed to generate a noise reduction network training sample pair s1 and S2, providing training samples for the noise reduction autoencoder; the sampling data is saved, the noise reduction data is received and output to the display device or computer. (3) calling the neural network processor. In the pre-sampling stage, the neural network processor is controlled to use the training samples to train the noise reduction autoencoder; in the formal sampling stage, the neural network processor is called to perform noise reduction processing on the vibration signal.

[0010] The neural network processor (NPU) is written with the network structure of the blind noise autoencoder. Its main components include a training module responsible for training the neural network and a prediction module responsible for noise reduction processing. In the pre-sampling stage, the training module uses the training samples input from the digital signal processing chip to train the network model, so that the noise reduction network is suitable for the noise reduction processing of the vibration signal of the current working condition, and the data network parameters will be saved to the prediction module. In the formal sampling stage, the noise reduction autoencoder in the prediction module is called to perform noise reduction processing on the input original vibration signal, and the processed noise reduction signal is output to the digital signal processing chip. Among them, the encoding layer of the blind noise reduction autoencoder is composed of three convolutional layers, and each convolutional layer is connected to the pooling layer. The decoding layer uses three upsampling pooling layers connected to three convolutional layers to complete decoding, and finally connects to the fully connected layer to output the results. The convolutional layers all use the ReLU activation function. The input length of the neural network is n, and the dimensions of the output data of each layer and the structure of the neural network are as follows. Figure 3 shown.

[0011] JTAG interface facilitates programming and debugging of various components within the system.

[0012] Another object of the present invention is to provide a vibration signal acquisition method with automatic noise reduction, using the vibration signal acquisition system with automatic noise reduction in the above embodiment, the vibration signal acquisition method with automatic noise reduction includes the following steps:

[0013] Step 1: Pre-collect signal. In the pre-collection mode, the clock chip, DSP control circuit and CPLD generate a dual sampling clock signal corresponding to the sampling frequency. The CPLD outputs the dual sampling clock signal to the ADC digital-to-analog conversion module, which controls it to collect the vibration signal at the sensor end in a dual sampling manner: the ADC obtains signal data at uniform sampling points at the set sampling frequency. After (i∈1, 2, ..., n), the sampling data is supplemented again at interval Δt And transmit it to DSP, which generates training sample data s1 and s1. The specific steps of sample generation are: for each sampling point data and Randomly select one of the two groups to generate observation signal s1, and then use the remaining data to generate observation data s2. The process is as follows Figure 6 shown.

[0014] Step 2: Train the noise reduction network. The neural network processor (NPU) receives the training sample pairs s1 and s2 from the DSP as the neural network input and labels, respectively. The training module is then activated to train the noise reduction autoencoder. The trained network parameters are saved to the prediction module, facilitating rapid noise reduction processing of the vibration signal during actual acquisition.

[0015] Step 3, formal data collection. During the formal data collection phase, the CPLD can be controlled to turn off double sampling, and the ADC sampling module can collect vibration signals normally by uniform sampling at the set sampling frequency. The DSP inputs the collected raw data s into the prediction module of the NPU for noise reduction processing to obtain the noise reduction signal.

[0016] The beneficial effects of the present invention are:

[0017] (1) The automatic noise reduction acquisition system proposed in this invention has a simple noise reduction process and does not require prior signal information or manual noise reduction parameter setting. This overcomes the limitation of existing noise reduction autoencoders that require noise-free signals as training labels. The training sample pairs obtained using only the double sampling method can meet the requirements for training the noise reduction autoencoder. After training, the neural network can quickly perform noise reduction on the collected vibration signals.

[0018] (2) The noise reduction method of the present invention is highly adaptable. Common noise reduction methods in engineering detection usually require a certain understanding of the prior information of the signal and select different noise reduction parameters to deal with signal noise under different working conditions. The noise reduction autoencoder based on deep learning proposed in the present invention can automatically learn signal features using pre-collected data, and does not require adjustment of the network structure for vibration signals under different working conditions. It can be widely used in noise reduction processing of various vibration signals and has good prospects for engineering application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is a hardware structure diagram of a vibration signal acquisition system with automatic noise reduction.

[0021] Figure 2 It is an operating process of a vibration signal acquisition system with automatic noise reduction.

[0022] Figure 3 This is a schematic diagram of the denoising autoencoder structure.

[0023] Figure 4 This is a schematic diagram of a dual sampling clock signal generation circuit.

[0024] Figure 5 This is a schematic diagram of the effects of uniform sampling and double sampling.

[0025] Figure 6 It is a schematic diagram of the training sample pair generation method.

[0026] Figure 7 This is the time domain diagram of the vibration signal in the case, (a) is the noise-free signal and (b) is the noisy signal.

[0027] Figure 8 is the training sample pair of the denoising autoencoder in this case, (a) is the training sample s1 and (b) is the training sample s2.

[0028] Figure 9 is the original noisy signal s and the noise-reduced signal after processing in this case Time domain diagram. DETAILED DESCRIPTION

[0029] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0030] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0031] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0032] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0033] Rolling bearing damage is one of the common factors that lead to mechanical failures. Accurate and efficient bearing damage monitoring methods are necessary to ensure the stable operation of engineering machinery. However, bearing vibration signals are usually non-stationary signals, and a large amount of noise is easily mixed in the signals. Therefore, noise reduction of noisy signals is an important part of bearing damage analysis. This case uses the data set provided by the Bearing Data Center of Case Western Reserve University (CWRU) as noise-free data to simulate the operation process of the present invention. Based on high-frequency data, a downsampling method with time intervals is used to simulate the vibration signal obtained by the double sampling method. Assuming that the original data is noise-free data, Gaussian white noise with a signal-to-noise ratio (SNR) of -5dB is added to it, and the noisy signal after noise addition is used to simulate the signal collected in the actual detection noise environment. The noise-free signal and the noisy signal are as follows. Figure 7 shown.

[0034] Step 1: Pre-collect signal. The noisy signal data after the noise is added is downsampled to simulate the double sampling process. On the basis of uniform sampling at a frequency of 16000 Hz, each sampling point is sampled again after an interval of 1 / 48000 second. The pre-collected double sampling signal data is then processed to generate the observation signals s1 and s2. The results are as follows: Figure 8 shown.

[0035] Step 2: Train the denoising network. The analog DSP uses two sets of signals, s1 and s2, as training sample pairs for the denoising autoencoder. These are input to the training module of the denoising chip's NPU to train the parameters of its neural network structure. The denoising autoencoder used in this case has an input dimension of (4000, 512). The optimization algorithm uses the adaptive moment estimation method (Adam), the loss function is the mean squared error (MSE), the learning rate is set to 0.0001, and the number of model training iterations is set to 100. After training is complete, the prediction module updates the parameters of its built-in network structure and uses it as the denoising network for actual acquisition.

[0036] Step 3: Formal acquisition. From the high-frequency signal data with -5dB Gaussian white noise added, extract the vibration signal data with a time step of 1 / 16000 seconds, simulate the noisy signal s obtained by uniform sampling of the vibration sensor at a sampling frequency of 16000Hz, and input it into the prediction module of the noise reduction chip to obtain the noise-reduced signal. The original noisy signal s and the denoised signal after denoising like Figure 9 shown.

[0037] In order to evaluate the noise reduction effect of this noise reduction method, the noisy signal and the noise reduction signal in this case were used for quantitative analysis, and the signal-to-noise ratio (SNR) and the root mean square error (RMSE) were calculated respectively compared with the noise-free signal. The signal-to-noise ratio of the signal was improved from -5.058dB to 3.681dB, indicating that the noise component was effectively suppressed; at the same time, the root mean square error of the signal was reduced from 0.4263 to 0.1559, and the noise reduction ratio was about 63.43%. It can be seen that the double sampling method proposed in the present invention can meet the requirements of the noise reduction autoencoder for training sample pairs. The blind noise reduction autoencoder constructed by this method can complete the training without using a noise-free signal, and has a good adaptive noise reduction effect on the collected vibration signal. This case proves the effectiveness of the noise reduction method and the feasibility of the vibration signal acquisition system with automatic noise reduction.

[0038] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A vibration signal acquisition method with automatic noise reduction, characterized in that: The vibration signal acquisition method with automatic noise reduction adopts a vibration signal acquisition system with automatic noise reduction, and the vibration signal acquisition method with automatic noise reduction includes the following steps: Step 1: Pre-collect signal; in the pre-collection mode, the clock chip, DSP control circuit and CPLD generate a dual sampling clock signal corresponding to the sampling frequency; the CPLD outputs the dual sampling clock signal to the ADC digital-to-analog conversion module, which controls it to collect the vibration signal at the sensor end in a dual sampling manner: the ADC obtains signal data at uniform sampling points at the set sampling frequency. After that, the sampling data is supplemented again at interval Δt And transmit it to DSP, which generates training sample data s1 and s2; Step 2: Train the noise reduction network. The neural network processor (NPU) receives the training sample pairs s1 and s2 from the DSP as the input and label of the neural network, respectively. The training module is activated to train the noise reduction autoencoder. The trained network parameters are saved to the prediction module to facilitate rapid noise reduction processing of the vibration signal during formal acquisition. Step 3, formal data collection; During the formal data collection stage, the CPLD can be controlled to turn off double sampling, and the ADC sampling module can collect vibration signals normally by uniform sampling at the set sampling frequency; the DSP inputs the collected raw data s into the prediction module of the NPU for noise reduction processing to obtain the noise reduction signal The vibration signal acquisition system for automatic noise reduction includes: a sensor module, a clock chip, an analog-to-digital converter, a programmable logic device, a digital signal processing chip, a neural network processor, and a JTAG interface. The sensor module is used to collect vibration signals and connect the vibration signals to the digital-to-analog converter in the form of analog signals; The clock chip is used to provide an operating clock signal for the system; The programmable logic device is electrically connected to the analog-to-digital converter and provides a sampling clock signal for the analog-to-digital converter; After receiving the sampling clock signal, the analog-to-digital converter converts the analog signal at the sensor end into a digital signal and transmits it to the digital signal processing chip; The digital signal processing chip is a microprocessor used for system control and data processing; The neural network processor is written with the network structure of the blind noise autoencoder; The JTAG interface is used for programming and debugging various components in the system.

2. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The sampling clock signal provided by the programmable logic device is uniform sampling or double sampling.

3. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The clock chip includes a crystal oscillation circuit and a charging circuit.

4. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The programmable logic device includes a control circuit, a counter and a pulse synthesis circuit.

5. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The neural network processor includes a training module responsible for training the neural network and a prediction module responsible for noise reduction processing.

6. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The encoding layer of the blind denoising autoencoder consists of three convolutional layers, each of which is connected to a pooling layer. The decoding layer uses three upsampling pooling layers connected to three convolutional layers to complete decoding, and finally connects to a fully connected layer to output the results. The convolutional layers all use the ReLU activation function.

7. The vibration signal acquisition method with automatic noise reduction according to claim 1, characterized in that: The specific steps of sample generation in step 1 include: and Randomly select one of the two groups to generate observation signal s1, and then use the remaining data to generate observation data s2.

Images