| The planetary reducer has the advantages of small volume,large transmission ratio,smooth transmission and high transmission efficiency.It is mainly used in the field of mechanical transmission of high speed with high power and low speed with high torque for example vehicles,ships and wind power generation.However,in the actual engineering applications,the planetary reducer often suffer from the complex dynamic force,It’s easy to occur mechanical failure.When the planetary reducer is faulty,if not found and dealt with timely,ranging from failure to cause the entire power transmission chain failure,while the heavy lead to significant economic losses and casualties.But because the planetary reducer is more complicated than the normal fixed shaft reducer,the planetary gear in the rotation process not only rotation,but also revolution;sun gear also with a number of planetary gear contact;The position of the meshing point of the defective gear teeth is different from the distance of the sensor as the gear rotates,resulting in a variety of modulation of the testing vibration signal,these will increase the complexity of the vibration signal.So that the impact characteristics caused by gear failure are usually submerged by a variety of vibration components and noise.Therefore,the fault diagnosis technique of the fixed shaft reducer cannot be applied to the planetary reducer.Requires effective fault diagnosis technology to extract the impact signal from the fault in the planetary reducer.However,only the fault impulse signal extracted by this method cannot quickly and accurately determine what type of fault occurred in a gear tooth of a planetary reducer.The deep learning method can obtain more abstract high-level data representation by combining low-level data,thus discovering the difference between the different categories of data.The original vibration signal data as the input of the network,not only the calculation of a long time,and the classification of the correct rate is low,so we need to extract the effective time domain and frequency domain characteristics of the planetary reducer fault vibration signal in as the depth of learning network input,in order to achieve fast classification and identification of planetary reducer.This paper first introduces the particularity of the planetary reducer compared with the general fixed shaft reducer,including the vibration mechanism of the general gear,the structural characteristics of the planetary reducer,the influence of the transmission path on the test signal,and the complexity of the vibration signal.This paper gives a detailed theoretical derivation of the local fault vibration signal model of planetary reducer.Then,based on the optimized Morlet wavelet and kurtosis,a pulse signal extraction method is proposed,which optimizes the Morlet wavelet parameters by using Shannon entropy to obtain the optimal center frequency and bandwidth parameters quickly.And then the wavelet transform of the original vibration signal is carried out by the optimized Morlet wavelet.Through the analysis,We reveal that the pulse signal should be reconstructed by several wavelet coefficients of the feature scale.Because the kurtosis index is sensitive to the pulse signal,the kurtosis index is used to select the feature scale and obtain the corresponding wavelet coefficients.In addition,due to the decomposition of the wavelet coefficients may be seriously polluted by noise,the wavelet coefficients can be denoised by applying the soft threshold method,finally,the wavelet coefficients are used to reconstruct the pulse signal.So that the fault pulse signal can be extracted from the original signal and verified with the analog signal.In this paper,in view of the shortcomings of the activation function used in the process of current deep belief network back propagation,the improved Sigmoid activation function is proposed and the theoretical algorithm is used to derive the function not only to speed up the updating of the network parameters,but also to improve the accuracy of the data classification,and it was verified by the classification of the public handwritten fontsThrough the construction of the planetary reducer test bed and the sun gear fault signal acquisition,the original signal and its extracted pulse signal are extracted in time domain and frequency domain characteristics.These characteristics are used as the input of the deep belief network。Respectively,the planetary reducer under the same loads with different fault types classify and identify、different loads with the same fault type classify and identify.and different loads with the different fault type classify and identify,the identification accuracy rate can reach more than 95%.Through the original vibration signal of the gears of the planetary reducer is directly used as the input of the network.It is shown that the extracted fault vibration signal is classified as the object,not only the calculation time is fast,but also the classification recognition accuracy is high.For all of the above fault classification recognition scheme,compared with the traditional activation function,the fault diagnosis method based on the improved depth belief network proposed in this paper can more accurately and quickly diagnose the planetary reducer fault type. |
