Rotary table mechanical resonance detection method based on neural network
Through the detection method based on neural network, the resonance state of the turntable is identified by signal acquisition and neural network model, which solves the problem of traditional methods being susceptible to noise interference and high misjudgment rate, and achieves high accuracy and stability mechanical resonance detection.
Patent Information
- Application Number
- CN202411544255.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-14
AI Technical Summary
Traditional rotary table mechanical resonance detection methods are susceptible to noise interference, have a high misjudgment rate, and are difficult to accurately identify resonance phenomena, resulting in safety risks of equipment and personnel.
Using a neural network-based detection method, the amplitude and frequency curves of the angle deviation signal and current signal are obtained through the signal acquisition module, the turntable working parameters are set, the resonance frequency is determined, and the training samples are enhanced and trained through the neural network model, and the prediction label is output. to identify resonance.
It improves the accuracy and stability of the detection, can accurately identify the resonant state of the turntable, reduces the safety risks of equipment and personnel, and improves the testing efficiency and success rate.
Smart Images
Figure CN119437397A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of mechanical resonance detection methods, and in particular to a turntable mechanical resonance detection method based on a neural network. Background Art
[0002] The turntable is a device used for inertial testing and space motion posture simulation. It is widely used in aerospace, military, automation, and robotics. Mechanical resonance may occur when the equipment is tested, causing interruptions in the test process and casualties to testers. In order to ensure the control performance of the turntable, it is urgent to improve the accuracy of detecting the mechanical resonance of the turntable. The traditional method of detecting the mechanical resonance of the turntable generally relies on engineering experience, is easily interfered by signals such as noise, and has a high misjudgment rate. Summary of the invention
[0003] In view of the above technical problems, the present invention proposes a turntable mechanical resonance detection method based on a neural network, including setting a turntable working mode and running the turntable. The detection method includes the following steps: Step S1, saving the angle deviation signal and the current signal through a signal acquisition module, and obtaining the amplitude-frequency curves of the angle deviation signal and the current signal respectively.
[0004] Step S2, setting the operating parameters of the turntable through the parameter adjustment module to determine the resonant frequency of the turntable.
[0005] Step S3, acquiring signal tags, judging and acquiring normal operation tags and resonance occurrence tags through a preset resonance frequency threshold.
[0006] Step S4, performing enhancement processing on the training samples, wherein the training samples are data set samples of the normally operating tags and the resonant tags and the corresponding signals.
[0007] Step S5: input all training samples into the neural network model for training.
[0008] Step S6, output the label value predicted by the neural network model.
[0009] Furthermore, the turntable working mode includes at least one motion position mode of position sine wave, position square wave, position triangle wave, and position trapezoidal wave.
[0010] Furthermore, the local control software or hardware of the turntable of the signal acquisition module, when the turntable is in a closed-loop position state, performs corresponding motion modes, acceleration, constant speed, and deceleration motion by setting different working modes, target positions, rates, and acceleration parameters to accurately reach the specified target position; at the same time, it can parse the data sent by the turntable, display the position and speed indicators of the turntable in real time, and save them in a local offline file.
[0011] Furthermore, the amplitude-frequency curve is an "angle deviation signal amplitude-frequency curve" obtained by fast Fourier transform of the "time-angle deviation signal", the horizontal axis is the frequency, and the vertical axis is the angle deviation.
[0012] Furthermore, the resonance frequency threshold is the frequency near the maximum peak in the amplitude-frequency curve.
[0013] Furthermore, the "time-current" test data is saved through the signal acquisition module, and normal operation labels and resonance labels are produced respectively. The normal operation label is "current-normal", where "normal" is digitized as 0; the resonance label is "current-resonance", where "resonance" is digitized as 1.
[0014] Furthermore, the label predicted by the output model is 0 or 1, where 0 represents "normal" and 1 represents "resonance occurs".
[0015] Furthermore, the signal enhancement is to perform non-uniform interval sampling on the training samples to obtain time-series resonance data samples in different sampling intervals.
[0016] Furthermore, the neural network model includes verification of the correctness of training samples, including multiple verification mechanisms, which review the training samples from different angles, including data accuracy, data consistency, and label accuracy.
[0017] The beneficial effects of the present invention compared with the prior art are: the detection method of the present invention has the advantages of high accuracy and high stability, which helps to accurately identify when the turntable resonates, avoid danger to equipment and personnel due to resonance, and at the same time ensure the test efficiency and success rate of the turntable. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 This is an example of the resonance detection method of the present invention. DETAILED DESCRIPTION
[0019] Example: Figure 1 A neural network-based turntable mechanical resonance detection method is shown, including setting the turntable working mode and running it. The detection method includes the following steps: Step S1, saving the angle deviation signal and the current signal through a signal acquisition module, and obtaining the amplitude-frequency curves of the angle deviation signal and the current signal respectively.
[0020] Step S2, setting the operating parameters of the turntable through the parameter adjustment module to determine the resonant frequency of the turntable.
[0021] Step S3, acquiring signal tags, judging and acquiring normal operation tags and resonance occurrence tags through a preset resonance frequency threshold.
[0022] Step S4, performing enhancement processing on the training samples, where the training samples are data set samples of the normally operating tags and the resonant tags and the corresponding signals.
[0023] Step S5: input all training samples into the neural network model for training.
[0024] Step S6, output the label value predicted by the neural network model.
[0025] The turntable working mode includes at least one motion position mode of position sine wave, position square wave, position triangle wave and position trapezoidal wave.
[0026] In this embodiment, in the position sine wave mode, the motion trajectory of the turntable presents a smooth sinusoidal curve change, which has the characteristics of periodicity and regularity, and can simulate some scenes with periodic motion requirements. The position square wave mode enables the turntable to quickly switch between two different position states, forming a square wave-like motion pattern, which is suitable for specific applications that require rapid position changes. In the position triangle wave mode, the motion trajectory of the turntable is triangular, and its speed change is relatively uniform, which can be used in some occasions with specific requirements for speed changes. In the position trapezoidal wave mode, the movement of the turntable presents a trapezoidal speed change curve, with different stages such as acceleration, uniform speed and deceleration, which can meet the needs of some complex motion control.
[0027] The signal acquisition module is the local control software or hardware of the turntable. When the turntable is in a closed-loop position state, it can perform corresponding motion modes, acceleration, constant speed, and deceleration by setting different working modes, target positions, speeds, and acceleration parameters to accurately reach the specified target position. At the same time, it can parse the data sent by the turntable, display the position and speed indicators of the turntable in real time, and save them in local offline files.
[0028] The amplitude-frequency curve is the "angle deviation signal amplitude-frequency curve" obtained by fast Fourier transform of the "time-angle deviation signal", where the horizontal axis is the frequency and the vertical axis is the angle deviation.
[0029] In this curve, the horizontal axis represents the frequency, which clearly shows the distribution of different frequency values; the vertical axis represents the angle deviation, which accurately reflects the degree of change of the angle deviation at different frequencies. The determination of the resonant frequency threshold is particularly critical, which is the frequency near the maximum peak in the amplitude-frequency curve. By accurately judging this frequency, it is possible to effectively identify whether the system is in a resonant state.
[0030] The resonant frequency threshold is the frequency near the maximum peak in the amplitude-frequency curve.
[0031] The signal acquisition module is used to save the "time-current" test data. In this process, normal operation labels and resonance labels are carefully made. The normal operation label is marked as "current-normal", and the "normal" state is digitally represented as 0, which represents that the system is in a stable and normal operation state without abnormalities. The resonance label is marked as "current-resonance", and the "resonance" state is digitized as 1. This number symbolizes that the system has a resonance phenomenon and is in a state that requires special attention and processing. The label predicted by the output model is 0 or 1, where 0 represents "normal" and 1 represents "resonance occurs".
[0032] Signal enhancement is to perform non-uniform sampling on the training samples to obtain time-series resonance data samples in different sampling intervals.
[0033] The neural network model includes verification of the correctness of training samples, including multiple verification mechanisms, which review the training samples from different angles, including data accuracy, data consistency, and label accuracy.
[0034] In this embodiment, first, the clearly marked normal operation labels and resonant operation labels are input into the carefully constructed neural network model. After that, the current signal is carefully enhanced. The data enhancement method of non-equally spaced sampling can effectively obtain richer and larger-scale resonant data samples. This processing method allows the model to be exposed to more resonant features in different situations, thereby improving the generalization ability of the model.
[0035] Next, the enhanced data is put into a neural network structure consisting of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer for comprehensive model training. In this process, the model hyperparameter learning rate is set to 0.01. This value has been carefully adjusted to ensure the training speed while avoiding falling into the local optimal solution. The number of training rounds is 1,000. Through such a large number of iterative training, the model can fully learn the characteristics and rules in the data and continuously optimize its own parameters to achieve better performance.
[0036] Finally, after the training process, the model will output accurate prediction labels. This prediction label will provide an important basis for the resonance detection of the turntable machinery. The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.
Claims
1. A method for detecting mechanical resonance of a turntable based on a neural network, comprising setting a turntable working mode and running the turntable, characterized in that: The detection method comprises the following steps: Step S1, saving the angle deviation signal and the current signal through a signal acquisition module, and obtaining the amplitude-frequency curves of the angle deviation signal and the current signal respectively; Step S2, setting the operating parameters of the turntable through the parameter adjustment module to determine the resonant frequency of the turntable; Step S3, obtaining signal tags, judging and obtaining normal operation tags and resonance tags by using a preset resonance frequency threshold; Step S4, performing enhancement processing on the training samples, wherein the training samples are data set samples of the normally operating tags and the resonant tags and the corresponding signals; Step S5, inputting all training samples into the neural network model for training; Step S6, output the label value predicted by the neural network model.
2. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The turntable working mode includes at least one motion position mode of position sine wave, position square wave, position triangle wave and position trapezoidal wave.
3. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The signal acquisition module is a local control software or hardware of the turntable. When the turntable is in a closed-loop position state, it can perform corresponding motion modes, acceleration, constant speed, and deceleration motion by setting different working modes, target positions, speeds, and acceleration parameters to accurately reach the specified target position. At the same time, it can parse the data sent by the turntable, display the position and speed indicators of the turntable in real time, and save them in a local offline file.
4. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The amplitude-frequency curve is the "angle deviation signal amplitude-frequency curve" obtained by fast Fourier transform of the "time-angle deviation signal", the horizontal axis is the frequency, and the vertical axis is the angle deviation.
5. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The resonance frequency threshold is the frequency near the maximum peak in the amplitude-frequency curve.
6. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The "time-current" test data is saved through the signal acquisition module, and normal operation labels and resonance labels are produced respectively. The normal operation label is "current-normal", where "normal" is digitized as 0; the resonance label is "current-resonance", where "resonance" is digitized as 1.
7. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The label predicted by the output model is 0 or 1, where 0 represents "normal" and 1 represents "resonance occurs".
8. The neural network-based turntable mechanical resonance detection method according to claim 1, characterized in that: The signal enhancement is to perform non-uniform interval sampling on the training samples to obtain time-series resonance data samples in different sampling intervals.
9. The method for detecting mechanical resonance of a turntable based on a neural network according to claim 1, characterized in that: The neural network model includes training sample correctness verification, including multiple verification mechanisms, and reviews the training samples from different angles, including data accuracy, data consistency, and label accuracy.