CN113536982A - Intelligent prediction method, system, equipment and medium for residual life of main reducer - Google Patents
Intelligent prediction method, system, equipment and medium for residual life of main reducer Download PDFInfo
- Publication number
- CN113536982A CN113536982A CN202110721641.6A CN202110721641A CN113536982A CN 113536982 A CN113536982 A CN 113536982A CN 202110721641 A CN202110721641 A CN 202110721641A CN 113536982 A CN113536982 A CN 113536982A
- Authority
- CN
- China
- Prior art keywords
- residual life
- feature extraction
- network
- monitoring data
- characteristic information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention relates to a method, a system, equipment and a medium for intelligently predicting the residual life of a main reducer, wherein the method comprises the steps of obtaining vibration data of a plurality of monitoring data channels of target equipment; respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information; respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information; inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model; and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy. The method can effectively detect the fault characteristics of the main speed reducer of the automobile and can accurately predict the residual service life of the main speed reducer of the automobile.
Description
Technical Field
The application relates to the technical field of mechanical equipment service life prediction, in particular to a method, a system, equipment and a medium for intelligently predicting the residual service life of a main speed reducer.
Background
The main reducer is used as a key part for reducing the rotating speed and increasing the torque in a rear axle of an automobile transmission system, is a main source of automobile faults, and has important influences on the working performance of the whole transmission system and the safety, comfort and reliability of an automobile due to the running state and the failure degree of the main reducer. Therefore, in order to ensure safe and reliable running of the vehicle, real-time state detection and residual life prediction of the automobile main speed reducer are necessary.
At present, the real-time state monitoring and the residual life prediction of an automobile main reducer are mainly realized by two modes, one mode is to simulate the recession trend of equipment by constructing a model of mechanical equipment, the method needs to fully understand the working principle of the equipment and the internal mechanical principle of the equipment, and the method has great challenge to scientific researchers; the other method is to analyze sensor monitoring data through various intelligent methods such as machine learning and the like and construct an intelligent prediction model based on data driving, and due to the limitation of technical means, the intelligent prediction technology still has the following problems to be solved urgently:
first, a strong noise environment is created due to the variation of load during the driving of the vehicle, the vibration interference of other components in the transmission system, and the sudden noise generated when the vehicle encounters an obstacle, and it is difficult to obtain an effective vibration signal. The abrasion degradation process of the main speed reducer in a strong noise environment has the characteristics of strong nonlinearity, non-stability and the like, and an ideal effect is difficult to obtain by directly predicting the vibration acceleration signals acquired by the sensor.
Secondly, because a single sensor is limited by the placement position and the direction, the multi-channel time sequence data acquired by a plurality of sensors can effectively reduce the uncertainty of single-source data, but the introduction of the heterogeneity of multi-sensor data can not be avoided, and the difficulty of deep time sequence feature extraction is increased. The wear failure of an automotive final drive as a typical life-time element is a process that gradually degrades over time, essentially a time series. The traditional technology has insufficient extraction capability of time sequence characteristics in the sensing signals, and important characteristics are easily lost.
Thirdly, in practical application, due to the complex running environment of the automobile and the randomness of the load working condition, the residual service life is a random variable, the residual service life of the main reducer and the running time have a complex nonlinear relation, and the discontinuous output of the traditional prediction model can cause low prediction accuracy.
In conclusion, the inventor believes that the existing intelligent prediction technology for the operation failure of the main speed reducer of the automobile has further room for improvement.
Disclosure of Invention
In order to overcome the defects of the prior art, the application provides the intelligent prediction method, the system, the equipment and the medium for the residual life of the main speed reducer, which can effectively detect the fault characteristics of the main speed reducer of the automobile and can more accurately predict the residual life of the main speed reducer of the automobile.
In a first aspect, the present application provides a method for intelligently predicting a remaining life of a main reducer, where the method includes:
obtaining vibration data of a plurality of monitoring data channels of target equipment;
respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information;
inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
Optionally, before the local feature extraction is performed on the vibration data of each monitoring data channel by using a preset multi-size convolutional neural network, the method further includes constructing a multi-size convolutional neural network, where the multi-size convolutional neural network includes a shallow feature extraction network, a local feature extraction network, and a preprocessing network, and the local feature extraction network and the preprocessing network are parallel networks behind the shallow feature extraction network;
the shallow feature extraction network adopts a single convolutional layer; the local feature extraction network comprises feature extraction layers with different convolution kernel sizes, and each feature extraction layer comprises a plurality of convolution layers, a plurality of activation layers and a pooling layer; the preprocessing network includes an active layer and a pooling layer.
Optionally, the step of performing local feature extraction on the vibration data of each monitoring data channel by using a preset multi-size convolutional neural network to obtain corresponding device failure feature information includes:
respectively inputting the vibration data of each monitoring data channel into one multi-size convolutional neural network for feature extraction by adopting a plurality of multi-size convolutional neural networks, and extracting the vibration data based on a shallow feature extraction network and a local feature extraction network to obtain local feature information; extracting the vibration data based on a shallow feature extraction network and a preprocessing network to obtain shallow target feature information; and obtaining equipment failure characteristic information based on fusion of the local characteristic information and the shallow layer target characteristic information.
Optionally, the deep feature extraction network adopts a bidirectional long-term and short-term memory network, and extracts the device failure feature information of each monitoring data channel by using a preset deep feature extraction network to obtain corresponding deep timing sequence feature information, including:
and a plurality of bidirectional long and short term memory networks are adopted, the equipment failure characteristic information of each monitoring data channel is respectively input into one bidirectional long and short term memory network, each bidirectional long and short term memory network simultaneously and bidirectionally processes the long sequence equipment failure characteristic information at a plurality of moments so as to obtain deep sequence characteristic information, and the deep sequence characteristic information represents the characteristic information of equipment failure and residual life change.
Optionally, before the deep timing characteristic information of the multiple monitoring data channels is input to a preset initial model for predicting remaining life for training, the method further includes: constructing a residual life prediction initial model, wherein the residual life prediction initial model adopts a deep neural network, the deep neural network comprises an input layer, a plurality of hidden layers and an output layer, the hidden layers of each layer adopt a full connection mode, and each hidden layer adopts a preset activation function; the neuron number settings of the hidden layer decrease in order from the input layer to the output layer.
Optionally, the deep timing characteristic information of the multiple monitoring data channels is input into a preset residual life prediction initial model for training, so as to obtain a residual life prediction target model, including:
and updating and training the connection weight of the residual life prediction initial model based on a back propagation algorithm by taking the mean square error between the residual life prediction value and the actual residual life as a loss function, and taking the trained residual life prediction initial model as a residual life prediction target model.
Optionally, predicting the remaining life of the target device at the current time by using a remaining life prediction target model based on a linear regression smoothing strategy, including:
and performing linear regression on the model prediction results of a plurality of moments before the current moment of the target equipment by using the residual life prediction target model, smoothing the model output of the current moment, and taking the linear regression result as the residual life of the current moment of the target equipment.
In a second aspect, the present application provides a system for intelligently predicting the remaining life of a final drive, the system comprising:
the data acquisition module is used for acquiring vibration data of a plurality of monitoring data channels of the target equipment;
the local feature extraction module is used for respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
the depth feature extraction module is used for extracting the equipment failure feature information of each monitoring data channel by using a preset deep feature extraction network to obtain corresponding deep time sequence feature information;
the model training module is used for inputting deep time sequence characteristic information of the monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
and the prediction module is used for predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
In a third aspect, the present application provides a computer device, which adopts the following technical solution:
a computer apparatus comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the intelligent prediction method of the final drive remaining life when executing the computer program.
In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solutions:
a computer-readable storage medium storing a computer program which, when executed by a processor, implements the steps of the intelligent prediction method of final drive remaining life.
The application has the following beneficial technical effects: the vibration data of a plurality of monitoring data channels are subjected to local feature extraction by respectively adopting a multi-size convolutional neural network, so that equipment failure feature information is effectively obtained, the equipment recession trend is conveniently reflected, and meanwhile, the over-fitting phenomenon can be effectively avoided; the equipment failure characteristic information obtained by each monitoring data channel is respectively subjected to characteristic extraction by adopting a deep characteristic extraction network, so that the nonlinear mapping relation between the vibration data and the residual service life is conveniently captured, and the corresponding deep time sequence characteristic information is obtained, thereby solving the problem of multi-channel data fusion and eliminating the heterogeneity existing between multi-channel time sequence data; inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training so as to obtain a residual life prediction target model; by adopting a linear regression smoothing strategy, the problem of discontinuity of the output value of the residual life prediction target model can be solved, so that the residual life of the target equipment can be predicted in real time, and the health management of the service life of the automobile main reducer is realized.
Drawings
FIG. 1 is a flow chart of a method of one embodiment of a method of mechanical fault monitoring of a vibration signal provided by the present invention;
FIG. 2 is a schematic diagram of data fusion for a multi-size convolutional neural network provided by the present invention;
FIG. 3 is a block diagram of a local feature extraction network provided by the present invention;
FIG. 4 is a schematic diagram of the multi-scale convolutional neural network provided by the present invention for processing multi-channel vibration data;
FIG. 5 is a schematic diagram of the multi-channel vibration data processing by the multi-scale convolutional neural network and the deep feature extraction network provided by the present invention;
FIG. 6 is a schematic diagram of the processing of the bidirectional long and short term memory network according to the present invention for local fusion features at multiple time instants;
FIG. 7 is a block diagram of a deep neural network provided by the present invention;
FIG. 8 is a schematic diagram of a linear regression smoothing strategy based on multiple time-point predictors according to the present invention;
FIG. 9 is a functional block diagram of one embodiment of a vibration signal mechanical fault monitoring system provided by the present invention.
Detailed Description
The present application is described in further detail below with reference to figures 1-9.
The embodiment of the application discloses a method for intelligently predicting the residual life of a main reducer, and with reference to fig. 1, the method comprises the following steps:
s1: obtaining vibration data of a plurality of monitoring data channels of target equipment;
s2: respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
s3: respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information;
s4: inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
s5: and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
In this embodiment, the target device refers to an automobile main reducer monitored by using the monitoring method of this embodiment, and in other embodiments, the target device may also be other mechanical devices requiring fault monitoring; the vibration condition of the automobile main speed reducer is monitored by adopting a plurality of acceleration sensors so as to obtain vibration data of a plurality of monitoring data channels, wherein the vibration data refer to vibration acceleration signals of the automobile main speed reducer.
The embodiment adopts the vibration data of a plurality of monitoring data channels, which is beneficial to improving the accuracy of the residual life prediction result; the vibration data of the monitoring data channels are subjected to local feature extraction by respectively adopting a multi-size convolutional neural network, so that the failure feature information of the equipment is effectively obtained, the equipment recession trend is conveniently reflected, and meanwhile, the over-fitting phenomenon can be effectively avoided; the equipment failure characteristic information obtained by each monitoring data channel is respectively subjected to characteristic extraction by adopting a deep characteristic extraction network, so that the nonlinear mapping relation between the vibration data and the residual service life is conveniently captured, and the corresponding deep time sequence characteristic information is obtained, thereby solving the problem of multi-channel data fusion and effectively eliminating the heterogeneity existing between multi-channel time sequence data; inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training so as to obtain a residual life prediction target model; by adopting a linear regression smoothing strategy, the problem of discontinuity of the output value of the residual life prediction target model can be solved, so that the residual life of the target equipment can be predicted in real time, and the health management of the service life of the automobile main reducer is realized.
In this embodiment, before the step S1 of using a preset multi-size convolutional neural network to respectively perform local feature extraction on the vibration data of each monitoring data channel, the method for monitoring a mechanical fault in this embodiment further includes constructing a multi-size convolutional neural network, where the multi-size convolutional neural network includes a shallow feature extraction network, a local feature extraction network, and a preprocessing network; referring to fig. 2, a local feature extraction network and a preprocessing network are used as parallel networks behind a shallow feature extraction network, and the shallow feature extraction network adopts a single convolutional layer; the local feature extraction network comprises feature extraction layers with different convolution kernel sizes, and each feature extraction layer comprises a plurality of convolution layers, a plurality of activation layers and a pooling layer; the preprocessing network includes an active layer and a pooling layer.
It should be noted that, because the acquired vibration data has the characteristics of high quantization, high dimensionality, strong noise, nonlinearity, non-stationarity and the like, the shallow feature extraction network is arranged before the local feature extraction network, so that the vibration data of the complete life cycle of the main reducer of the automobile can be conveniently subjected to convolution processing, and the dimensionality, the stationarity and the noise level of the vibration acceleration signal are reduced. In the present embodiment, a single convolution layer of the shallow feature extraction network employs 1 × 4 convolution kernels, and the number of convolution kernels and the convolution step size are set to 64 and 1, respectively. The activation layer in the pre-processing network uses the ReLu activation function, and the pooling layer uses a max-pooling strategy.
Furthermore, the local feature extraction network adopts convolution kernels with different sizes, nonlinear transformation is carried out on a convolution result by combining a nonlinear activation function of an activation layer, and finally down-sampling processing is carried out by a pooling layer to reduce dimensionality, so that local feature components in vibration data can be extracted, the characteristics of a decline and failure mode are formed, and an over-fitting phenomenon can be effectively avoided. Referring to fig. 3, the local feature extraction network of the present embodiment includes three parallel feature extraction layers, the sizes of convolution kernels in the three layers are respectively 3 × 3, 5 × 5, and 7 × 7, each convolution processing is followed by an active layer processing based on a ReLu function, and a pooling layer down-samples the output of the previous layer using max-posing; the number of convolution kernels is set to 64, i.e., the output of each feature extraction layer is 64 feature maps.
In this embodiment, step S1 is to perform local feature extraction on the vibration data of each monitoring data channel by using a preset multi-size convolutional neural network, so as to obtain corresponding device failure feature information, and the method includes the following steps:
adopting a plurality of multi-size convolutional neural networks, respectively inputting the vibration data of each monitoring data channel into one multi-size convolutional neural network for feature extraction, and extracting the vibration data based on a shallow feature extraction network and a local feature extraction network to obtain local feature information; extracting the vibration data based on a shallow feature extraction network and a preprocessing network to obtain shallow target feature information; and obtaining equipment failure characteristic information based on fusion of the local characteristic information and the shallow layer target characteristic information.
It should be noted that, referring to fig. 2 and 4, a feature-level data fusion technique is adopted to fuse shallow target feature information and local feature information of vibration data, in this embodiment, a "jump connection" of a residual network is introduced into a feature fusion process to perform jump-type merging of the shallow feature and the local feature, so as to obtain a local fusion feature, that is, device failure feature information, thereby effectively avoiding a gradient disappearance phenomenon caused by too many network layers and contributing to improving the prediction accuracy of the subsequent remaining life; it should be noted that, the residual network of this embodiment adopts a ResNet network, and in the training process of the residual network (ResNet), the problems of gradient disappearance and gradient explosion that easily occur in the deep neural network are solved by using two technologies, namely, a residual block and a "jump connection". The term "jump connection" refers to a process in which the output of a certain layer jumps over the next layer or a plurality of subsequent layers and information is transferred to a neural network at a deeper layer in a jumping manner.
In this embodiment, in step S2, the deep feature extraction network adopts a Bidirectional long short term memory network (bilastm), and the preset deep feature extraction network is used to extract the device failure feature information of each monitored data channel, so as to obtain corresponding deep timing feature information, including the following steps:
and a plurality of bidirectional long and short term memory networks are adopted, the equipment failure characteristic information of each monitoring data channel is respectively input into one bidirectional long and short term memory network, each bidirectional long and short term memory network simultaneously and bidirectionally processes the long sequence equipment failure characteristic information at a plurality of moments so as to obtain deep sequence characteristic information, and the deep sequence characteristic information represents the characteristic information of equipment failure and residual life change.
It should be noted that the vibration data acquired by the acceleration sensor belongs to time sequence data, and when the change rule of the time sequence data is analyzed, the two-way long-short term memory network can be used for capturing the long-time dependency relationship, namely the deep time sequence characteristic information of the vibration data; after the vibration data are processed by a multi-size convolution neural network, local fusion characteristic information, namely equipment failure characteristic information is obtained; referring to fig. 5, local fusion characteristic information of a plurality of monitoring data channels is used as input data of a corresponding number of bidirectional long and short term memory networks, a nonlinear mapping relation between vibration data and the residual life is captured through processing of a memory body and a door structure of the bidirectional long and short term memory networks, characteristic information related to time, namely deep time sequence characteristic information, is obtained through learning, and information irrelevant to equipment decline can be effectively filtered; referring to fig. 6, in the present embodiment, the bidirectional long-short term memory network is adopted to fully utilize the influence of the feature information of the past time (t-1) and the future time (t +1) on the target time (t), and simultaneously bidirectionally process the long sequence data of multiple times, thereby solving the problem of long-term memory dependence, being capable of better learning deep sequence feature information in the time dimension, and reflecting the change of the equipment failure process and the remaining life. In addition, in other embodiments, an automatic encoder or a deep neural network may be employed to achieve deep feature mining of the multi-channel data.
In this embodiment, before inputting Deep timing characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training in step S3, the mechanical fault monitoring method of this embodiment further includes constructing the residual life prediction initial model, where the residual life prediction initial model employs a Deep Neural Network (DNN), the Deep Neural network includes an input layer, a plurality of hidden layers, and an output layer, where all hidden layers are connected in a full-connection manner, and each hidden layer employs a preset activation function; the neuron number settings of the hidden layer decrease in order from the input layer to the output layer.
It should be noted that, referring to fig. 7, the deep neural network of this embodiment adopts 6 hidden layers, the number of neurons in each layer is sequentially set to 200, 100, 50, 30, 10, and 1, respectively, and each hidden layer contains a ReLu activation function, so as to implement nonlinear transformation on input features. In other embodiments, the number of the hidden layers and the number of the neurons can be adaptively set according to actual needs.
In this embodiment, step S3 is to input deep timing characteristic information of multiple monitoring data channels into a preset initial model for remaining life prediction for training, so as to obtain a target model for remaining life prediction, including the following steps:
and taking the mean square error between the predicted value of the residual life and the actual residual life as a loss function, updating and training the connection weight of the initial residual life prediction model based on a back propagation algorithm, and taking the trained initial residual life prediction model as a residual life prediction target model.
It should be noted that the deep timing characteristic information of all the monitoring data channels is used as training data, the training data is used as input of the deep neural network, the predicted value of the remaining life is used as output, and the connection weight of the full connection layer of the deep neural network is updated through a back propagation algorithm of mean square error.
In this embodiment, the step S4 of predicting the remaining life of the target device at the current time by using the remaining life prediction target model based on the linear regression smoothing strategy includes the following steps:
and performing linear regression on the model prediction results of a plurality of moments before the current moment of the target equipment by using the residual life prediction target model, smoothing the model output of the current moment, and taking the linear regression result as the residual life of the current moment of the target equipment.
It should be noted that the output of the remaining life prediction target model is usually expressed asAccording to the characteristics of discontinuity, fluctuation and the like, the prediction results based on a plurality of earlier-stage moments are used for linear regression, so that the residual life prediction and the running time of the automobile main reducer are in a linear relation, for example, the residual life prediction target model has a RUL (residual life indicator) as the equipment residual life output value of the current moment ttBased on the output value [ RUL ] at time t and 8 time points before itt-7,RULt-6,RULt-5,…,RULt-1,RULt]Linear Regression (Linear Regression) is performed, and the output result of the Regression is a continuous value related to the operation time, that is, the final predicted value of the remaining life at the current time t, see fig. 8.
The vibration data of the monitoring data channels are respectively subjected to feature extraction by adopting a multi-size convolutional neural network, so that equipment failure feature information is effectively obtained, the equipment recession trend can be conveniently reflected, the over-fitting phenomenon can be avoided, and the shallow layer features and the local features are subjected to jumping fusion, so that the problems of gradient disappearance and gradient explosion caused by excessive network layers can be effectively solved, and the accuracy of feature extraction is improved; the device failure characteristic information of each monitoring data channel is subjected to deep characteristic extraction through a bidirectional long-short term memory network, so that the nonlinear mapping relation between the vibration data and the residual service life is conveniently captured, and deep time sequence characteristic information is mined, so that the problem of multi-channel data fusion is solved, and the heterogeneity existing among multi-channel time sequence data is eliminated; inputting deep time sequence characteristic information of all monitoring data channels into a deep neural network containing a plurality of fully-connected layers, and obtaining a residual life prediction target model through back propagation training; the problem of discontinuity of the output value of the prediction model is solved through a linear regression smoothing strategy of the predicted values at multiple moments, so that the residual life of the target equipment can be predicted in real time, and the health management of the service life of the automobile main reducer is realized.
It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
The embodiment also provides an intelligent prediction system for the residual life of the main reducer, and the intelligent prediction system for the residual life of the main reducer corresponds to the intelligent prediction method for the residual life of the main reducer in the embodiment one by one. As shown in fig. 9, the intelligent prediction system for the residual life of the main reducer comprises a data acquisition module 901, a local feature extraction module 902, a depth feature extraction module 903, a model training module 904 and a prediction module 905. The functional modules are explained in detail as follows:
a data obtaining module 901, configured to obtain vibration data of multiple monitoring data channels of a target device;
the local feature extraction module 902 is configured to perform local feature extraction on the vibration data of each monitoring data channel by using a preset multi-size convolutional neural network, so as to obtain corresponding device failure feature information;
a depth feature extraction module 903, configured to extract device failure feature information of each monitoring data channel by using a preset deep feature extraction network, so as to obtain corresponding deep timing feature information;
the model training module 904 is configured to input deep timing characteristic information of the multiple monitoring data channels into a preset residual life prediction initial model for training, so as to obtain a residual life prediction target model;
and the prediction module 905 is configured to predict the remaining life of the target device at the current time based on a linear regression smoothing strategy and by using the remaining life prediction target model.
For specific limitations of the intelligent prediction system for the residual life of the final drive unit, reference may be made to the above limitations of the intelligent prediction method for the residual life of the final drive unit, and details thereof are not repeated herein. All or part of each module in the intelligent prediction system for the residual life of the main speed reducer can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
The present embodiments also provide a computer device, which may be a server, comprising a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer equipment is used for storing vibration data, equipment failure characteristic information, deep time sequence characteristic information and the like of a plurality of monitoring data channels. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize the intelligent prediction method for the residual life of the main reducer, and the processor executes the computer program to realize the following steps:
s1: obtaining vibration data of a plurality of monitoring data channels of target equipment;
s2: respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
s3: respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information;
s4: inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
s5: and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
The present embodiments also provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:
s1: obtaining vibration data of a plurality of monitoring data channels of target equipment;
s2: respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
s3: respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information;
s4: inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
s5: and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions. The above embodiments are preferred embodiments of the present application, and the protection scope of the present application is not limited by the above embodiments, so: all equivalent changes made according to the structure, shape and principle of the present application shall be covered by the protection scope of the present application.
Claims (10)
1. The intelligent prediction method for the residual life of the main reducer is characterized by comprising the following steps: the method comprises the following steps:
obtaining vibration data of a plurality of monitoring data channels of target equipment;
respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
respectively extracting the equipment failure characteristic information of each monitoring data channel by using a preset deep characteristic extraction network to obtain corresponding deep time sequence characteristic information;
inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
and predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
2. The intelligent prediction method for the residual life of the main reducer according to claim 1, characterized in that: before the local feature extraction is performed on the vibration data of each monitoring data channel by using the preset multi-size convolutional neural network, the method further includes:
constructing a multi-size convolutional neural network, wherein the multi-size convolutional neural network comprises a shallow feature extraction network, a local feature extraction network and a preprocessing network, and the local feature extraction network and the preprocessing network are used as parallel networks behind the shallow feature extraction network;
the shallow feature extraction network adopts a single convolutional layer; the local feature extraction network comprises feature extraction layers with different convolution kernel sizes, and each feature extraction layer comprises a plurality of convolution layers, a plurality of activation layers and a pooling layer; the preprocessing network includes an active layer and a pooling layer.
3. The intelligent prediction method for the residual life of the main reducer according to claim 2, characterized in that: the method for extracting the local features of the vibration data of each monitoring data channel by utilizing the preset multi-size convolutional neural network to obtain the corresponding equipment failure feature information comprises the following steps:
respectively inputting the vibration data of each monitoring data channel into one multi-size convolutional neural network for feature extraction by adopting a plurality of multi-size convolutional neural networks, and extracting the vibration data based on a shallow feature extraction network and a local feature extraction network to obtain local feature information;
extracting the vibration data based on a shallow feature extraction network and a preprocessing network to obtain shallow target feature information; and obtaining equipment failure characteristic information based on fusion of the local characteristic information and the shallow layer target characteristic information.
4. The intelligent prediction method for the residual life of the main reducer according to claim 1, characterized in that: the deep layer feature extraction network adopts a bidirectional long-short term memory network, and extracts the equipment failure feature information of each monitoring data channel by utilizing a preset deep layer feature extraction network to obtain corresponding deep layer time sequence feature information, and the method comprises the following steps:
and a plurality of bidirectional long and short term memory networks are adopted, the equipment failure characteristic information of each monitoring data channel is respectively input into one bidirectional long and short term memory network, each bidirectional long and short term memory network simultaneously and bidirectionally processes the long sequence equipment failure characteristic information at a plurality of moments so as to obtain deep sequence characteristic information, and the deep sequence characteristic information represents the characteristic information of equipment failure and residual life change.
5. The intelligent prediction method for the residual life of the main reducer according to claim 1, characterized in that: before inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training, the method further comprises the following steps:
constructing a residual life prediction initial model, wherein the residual life prediction initial model adopts a deep neural network, the deep neural network comprises an input layer, a plurality of hidden layers and an output layer, the hidden layers of each layer adopt a full connection mode, and each hidden layer adopts a preset activation function;
the neuron number settings of the hidden layer decrease in order from the input layer to the output layer.
6. The intelligent prediction method for the residual life of the main reducer according to claim 5, characterized in that: inputting deep time sequence characteristic information of a plurality of monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model, wherein the residual life prediction target model comprises the following steps:
and updating and training the connection weight of the residual life prediction initial model based on a back propagation algorithm by taking the mean square error between the residual life prediction value and the actual residual life as a loss function, and taking the trained residual life prediction initial model as a residual life prediction target model.
7. The intelligent prediction method for the residual life of the main reducer according to claim 1, characterized in that: predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy, wherein the prediction comprises the following steps:
and performing linear regression on the model prediction results of a plurality of moments before the current moment of the target equipment by using the residual life prediction target model, smoothing the model output of the current moment, and taking the linear regression result as the residual life of the current moment of the target equipment.
8. The utility model provides a main reducer residual life intelligent prediction system which characterized in that: the system comprises:
the data acquisition module is used for acquiring vibration data of a plurality of monitoring data channels of the target equipment;
the local feature extraction module is used for respectively carrying out local feature extraction on the vibration data of each monitoring data channel by utilizing a preset multi-size convolutional neural network to obtain corresponding equipment failure feature information;
the depth feature extraction module is used for extracting the equipment failure feature information of each monitoring data channel by using a preset deep feature extraction network to obtain corresponding deep time sequence feature information;
the model training module is used for inputting deep time sequence characteristic information of the monitoring data channels into a preset residual life prediction initial model for training to obtain a residual life prediction target model;
and the prediction module is used for predicting the residual life of the target equipment at the current moment by using a residual life prediction target model based on a linear regression smoothing strategy.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor when executing the computer program implements the steps of the intelligent prediction method of the residual life of the final drive according to any one of claims 1 to 7.
10. A computer-readable storage medium, which stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the intelligent prediction method for remaining life of a final drive according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110721641.6A CN113536982B (en) | 2021-06-28 | 2021-06-28 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110721641.6A CN113536982B (en) | 2021-06-28 | 2021-06-28 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113536982A true CN113536982A (en) | 2021-10-22 |
| CN113536982B CN113536982B (en) | 2023-02-24 |
Family
ID=78097145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110721641.6A Active CN113536982B (en) | 2021-06-28 | 2021-06-28 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113536982B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282646A (en) * | 2021-11-29 | 2022-04-05 | 淮阴工学院 | Light power prediction method and system based on two-stage feature extraction and improved BilSTM |
| CN115330094A (en) * | 2022-10-14 | 2022-11-11 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipeline service life prediction method, internet of things system, device and medium |
| CN115879248A (en) * | 2023-03-03 | 2023-03-31 | 山东亿宁环保科技有限公司 | Full life cycle management method and system suitable for vacuum pump |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522600A (en) * | 2018-10-16 | 2019-03-26 | 浙江大学 | Complex equipment remaining life prediction technique based on combined depth neural network |
| CN109726524A (en) * | 2019-03-01 | 2019-05-07 | 哈尔滨理工大学 | A kind of rolling bearing remaining life prediction technique based on CNN and LSTM |
| CN110378052A (en) * | 2019-07-25 | 2019-10-25 | 北京航空航天大学 | It is looked to the future the equipment method for predicting residual useful life of operating condition based on Recognition with Recurrent Neural Network |
| CN112036547A (en) * | 2020-08-28 | 2020-12-04 | 江苏徐工信息技术股份有限公司 | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM |
| CN112257333A (en) * | 2020-09-24 | 2021-01-22 | 浙江工业大学 | Mechanical equipment internal component service life prediction method based on deep learning |
| CN112926273A (en) * | 2021-04-13 | 2021-06-08 | 中国人民解放军火箭军工程大学 | Method for predicting residual life of multivariate degradation equipment |
| WO2021120157A1 (en) * | 2019-12-20 | 2021-06-24 | Intel Corporation | Light weight multi-branch and multi-scale person re-identification |
-
2021
- 2021-06-28 CN CN202110721641.6A patent/CN113536982B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522600A (en) * | 2018-10-16 | 2019-03-26 | 浙江大学 | Complex equipment remaining life prediction technique based on combined depth neural network |
| CN109726524A (en) * | 2019-03-01 | 2019-05-07 | 哈尔滨理工大学 | A kind of rolling bearing remaining life prediction technique based on CNN and LSTM |
| CN110378052A (en) * | 2019-07-25 | 2019-10-25 | 北京航空航天大学 | It is looked to the future the equipment method for predicting residual useful life of operating condition based on Recognition with Recurrent Neural Network |
| WO2021120157A1 (en) * | 2019-12-20 | 2021-06-24 | Intel Corporation | Light weight multi-branch and multi-scale person re-identification |
| CN112036547A (en) * | 2020-08-28 | 2020-12-04 | 江苏徐工信息技术股份有限公司 | Rolling bearing residual life prediction method combining automatic feature extraction with LSTM |
| CN112257333A (en) * | 2020-09-24 | 2021-01-22 | 浙江工业大学 | Mechanical equipment internal component service life prediction method based on deep learning |
| CN112926273A (en) * | 2021-04-13 | 2021-06-08 | 中国人民解放军火箭军工程大学 | Method for predicting residual life of multivariate degradation equipment |
Non-Patent Citations (3)
| Title |
|---|
| QING YE ET AL.: "A Deep Learning Model for Fault Diagnosis with a Deep Neural Network and Feature Fusion on Multi-Channel Sensory Signals", 《SENSORS》 * |
| TANGBIN XIA ET AL.: "An ensemble framework based on convolutional bi-directional LSTM with multiple time windows for remaining useful life estimation", 《COMPUTERS IN INDUSTRY》 * |
| 叶青 等: "基于改进SBELM的耦合故障诊断方法", 《西南交通大学学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282646A (en) * | 2021-11-29 | 2022-04-05 | 淮阴工学院 | Light power prediction method and system based on two-stage feature extraction and improved BilSTM |
| CN114282646B (en) * | 2021-11-29 | 2023-08-25 | 淮阴工学院 | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement |
| CN115330094A (en) * | 2022-10-14 | 2022-11-11 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipeline service life prediction method, internet of things system, device and medium |
| US11898704B2 (en) | 2022-10-14 | 2024-02-13 | Chengdu Qinchuan Iot Technology Co., Ltd. | Methods and Internet of Things systems for smart gas pipeline life prediction based on safety |
| CN115879248A (en) * | 2023-03-03 | 2023-03-31 | 山东亿宁环保科技有限公司 | Full life cycle management method and system suitable for vacuum pump |
| CN115879248B (en) * | 2023-03-03 | 2023-05-16 | 山东亿宁环保科技有限公司 | Full life cycle management method and system suitable for vacuum pump |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113536982B (en) | 2023-02-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113536982B (en) | Intelligent prediction method, system, equipment and medium for residual life of main reducer | |
| Hanselmann et al. | CANet: An unsupervised intrusion detection system for high dimensional CAN bus data | |
| CN109581871B (en) | Industrial control system intrusion detection method of immune countermeasure sample | |
| Qin et al. | Anti‐noise diesel engine misfire diagnosis using a multi‐scale CNN‐LSTM neural network with denoising module | |
| Chen et al. | Embedded system real-time vehicle detection based on improved YOLO network | |
| WO2014018793A1 (en) | Apparatus and methods for efficient updates in spiking neuron networks | |
| CN112001314A (en) | Early fault detection method for variable speed hoist | |
| CN110736968A (en) | Radar abnormal state diagnosis method based on deep learning | |
| CN112882909A (en) | Fuel cell system fault prediction method and device | |
| CN112861071B (en) | High-speed rail traction system anomaly detection method based on depth self-coding | |
| CN110723152B (en) | Artificial intelligence detection method, device, equipment or storage medium for guaranteeing automatic driving safety | |
| CN114937461A (en) | Live pig sound event detection method and device based on channel attention and residual gating convolution | |
| CN113360339A (en) | Preprocessing method for providing uniform data blocks | |
| Lin et al. | Remaining useful life prediction in prognostics using multi-scale sequence and Long Short-Term Memory network⋆ | |
| CN115660198B (en) | Method for predicting residual service life of rolling bearing | |
| CN113850138B (en) | Unmanned aerial vehicle sensor fault detection method, system and device based on data driving | |
| CN112596388B (en) | LSTM neural network AEB system control method based on driver data | |
| KR20220108678A (en) | Arrhythmia classification method using long short-term memory based on recurrent neural network | |
| CN116226770A (en) | Time sequence data anomaly detection method and device | |
| CN115409217A (en) | Multitask predictive maintenance method based on multi-expert hybrid network | |
| CN115409262A (en) | Railway data center key performance index trend prediction method and abnormity identification method | |
| CN114810292A (en) | Computer-implemented method and apparatus for maneuver identification of exhaust aftertreatment systems | |
| CN113982605A (en) | Multi-level shield tunnel safety protection system and method | |
| CN112560252A (en) | Prediction method for residual life of aircraft engine | |
| KR102383411B1 (en) | Apparatus for adaptive predictive maintenance of industrial robot using dynamic classification signal learning and method thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |