CN113297704A - Harmonic reducer power real-time prediction method and system based on hybrid deep neural network - Google Patents
Harmonic reducer power real-time prediction method and system based on hybrid deep neural network Download PDFInfo
- Publication number
- CN113297704A CN113297704A CN202110672801.2A CN202110672801A CN113297704A CN 113297704 A CN113297704 A CN 113297704A CN 202110672801 A CN202110672801 A CN 202110672801A CN 113297704 A CN113297704 A CN 113297704A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deep neural
- harmonic reducer
- layer
- data
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- 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
Abstract
The invention provides a harmonic reducer power real-time prediction method and a system based on a hybrid deep neural network, which comprises the following steps: building a harmonic reducer accelerated life test platform, and collecting operation and monitoring parameters of the harmonic reducer; constructing a data set based on the collected operation and monitoring parameters, and preprocessing the data set; dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion; training the mixed deep neural network by using a training set to obtain a trained mixed deep neural network; predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, and adjusting the hyper-parameter of the trained mixed deep neural network when the prediction error does not meet the preset requirement until the prediction error meets the preset requirement; and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
Description
Technical Field
The invention relates to the technical field of harmonic reducers, in particular to a method and a system for predicting the power of a harmonic reducer in real time based on a hybrid deep neural network.
Background
The harmonic reducer is one of the core components of the industrial robot, has the remarkable advantages of high transmission efficiency, high transmission precision, large single-machine transmission ratio, small size, strong bearing capacity and the like, and is widely applied to the field of industrial robots. Because the harmonic reducer mainly plays the effect of reducing output rotating speed and increasing output torque, the performance of the harmonic reducer is directly related to the positioning precision and the health state of the industrial robot. The power signal of the harmonic reducer is one of the important parameters characterizing the operating state of the harmonic reducer. Under the long-time load work, the performance of the harmonic reducer is degraded, so that the power of the harmonic reducer is accurately predicted in real time, and the method has guiding significance for reducing the loss caused by sudden failure, improving the performance of the industrial robot and prolonging the service life of the industrial robot.
At present, in the research aiming at the fault detection and performance prediction of the harmonic reducer, most of the research is to calculate the performance degradation process of the harmonic reducer through theoretical formulas, but the theoretical formulas usually depend on preset conditions, and the accuracy and the practicability of the theoretical formulas are still to be improved. Meanwhile, most industrial mechanical arms do cyclic reciprocating motion, the load distribution and parameter change of each joint are nonlinear, and how to effectively utilize decline data of the harmonic reducer to carry out accurate and reliable fault detection and performance prediction on the industrial mechanical arms becomes a problem to be solved urgently. Aiming at the problems and the defects of related research work of harmonic reducer performance prediction, a CNN and BilSTM-based mixed deep neural network (DCBNN) is provided, and the operating power of the harmonic reducer is accurately predicted by using the operating and state parameters of the harmonic reducer acquired by an accelerated life test. Firstly, the first branch of the model utilizes a convolutional neural network to extract the spatial characteristics of the state monitoring data, and simultaneously, the other branch utilizes two layers of BilSTMs to effectively extract the time sequence characteristics of the state monitoring data, and the combination of the CNN and the BilSTM can effectively extract the deep space-time variation characteristics of the state monitoring data and alleviate the problem of gradient disappearance, thereby improving the accuracy and stability of power prediction. And finally, verifying the effectiveness of the method by utilizing a harmonic reducer accelerated life experimental data set.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a harmonic reducer power real-time prediction method and system based on a hybrid deep neural network.
The invention provides a harmonic reducer power real-time prediction method based on a hybrid deep neural network, which comprises the following steps:
step S1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
step S2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
step S3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
step S4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
step S5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
step S6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
Preferably, the acquiring operation and monitoring parameters of the harmonic reducer in a normal operation state includes: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
Preferably, the preprocessing the data set in the step S2 includes: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
Preferably, the hybrid deep neural network includes: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
Preferably, the step S4 includes: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
The invention provides a harmonic reducer power real-time prediction system based on a hybrid deep neural network, which comprises the following components:
module M1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
module M2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
module M3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
module M4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
module M5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
module M6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
Preferably, the acquiring operation and monitoring parameters of the harmonic reducer in a normal operation state includes: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
Preferably, the preprocessing the data set in the module M2 includes: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
Preferably, the hybrid deep neural network includes: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
Preferably, said module M4 comprises: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
Compared with the prior art, the invention has the following beneficial effects:
1. aiming at the nonlinear transient characteristic of the harmonic reducer, a hybrid deep neural network model is provided by utilizing multidimensional operation monitoring parameters to predict the power of the harmonic reducer, and the hybrid deep neural network model is utilized to automatically learn the required characteristic;
2. the hybrid deep neural network model comprehensively extracts the spatial coupling characteristics and the bidirectional time sequence dependency relationship of the monitoring data, and utilizes the operation parameters of the harmonic reducer more efficiently, so that the hybrid deep neural network model has higher model prediction precision and universality;
3. the method has important significance for predicting the power of the harmonic reducer in real time in practical application and further predicting the performance of the harmonic reducer.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a flow chart of a method for predicting power of a harmonic reducer in real time based on a hybrid deep neural network according to the invention;
FIG. 2 is a block diagram of a hybrid deep neural network of the present invention;
FIG. 3 is a flow chart of the hybrid deep neural network training of the present invention;
FIG. 4 is a diagram illustrating actual values of power in a test set according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating power prediction values on a test set according to an embodiment of the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
Example 1
The invention provides a harmonic reducer power real-time prediction method based on a hybrid deep neural network, which comprises the following steps:
step S1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
step S2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
step S3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
step S4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
step S5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
step S6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
Specifically, the acquiring operation and monitoring parameters of the harmonic reducer in a normal operation state includes: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
Specifically, the preprocessing the data set in step S2 includes: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
Specifically, the step S3 includes: setting a time domain window to be 20, dividing the operation and monitoring data of the harmonic reducer by using a sliding window method, and setting the proportion of a training set to a testing set to be 8:2 or 7:3 according to different data sets of the obtained set. Meanwhile, the training set is divided into a training set and a verification set according to a ratio of 9:1 before training begins.
Specifically, the hybrid deep neural network includes: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
Specifically, the step S4 includes: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
The invention provides a harmonic reducer power real-time prediction system based on a hybrid deep neural network, which comprises the following components:
module M1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
module M2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
module M3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
module M4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
module M5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
module M6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
Specifically, the acquiring operation and monitoring parameters of the harmonic reducer in a normal operation state includes: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
Specifically, the preprocessing the data set in the module M2 includes: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
Specifically, the module M3 includes: setting a time domain window to be 20, dividing the operation and monitoring data of the harmonic reducer by using a sliding window method, and setting the proportion of a training set to a testing set to be 8:2 or 7:3 according to different data sets of the obtained set. Meanwhile, the training set is divided into a training set and a verification set according to a ratio of 9:1 before training begins.
Specifically, the hybrid deep neural network includes: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
Specifically, the module M4 includes: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
Example 2
Example 2 is a preferred example of example 1
The invention discloses a harmonic reducer power real-time prediction method based on a hybrid deep neural network, which is used for processing state monitoring data of a harmonic reducer so as to accurately predict a power signal of the harmonic reducer; because the load distribution and parameter change of the harmonic reducer are nonlinear, the running power of the harmonic reducer needs to be accurately predicted in real time, and then fault detection and performance prediction of the harmonic reducer are guided; carrying out data preprocessing on the collected operating parameters of the harmonic reducer, and dividing a data set; extracting spatial characteristics and bidirectional time sequence dependence of state monitoring data by utilizing a convolutional neural network and a bidirectional long-short term memory neural network branch; and training by using different harmonic reducer state detection data sets, and predicting the power of the test set by using the trained hybrid model.
As shown in fig. 1, the method for predicting the power of a harmonic reducer in real time based on a hybrid deep neural network provided by the invention comprises the following steps:
step 1: constructing a harmonic reducer accelerated life experiment platform, collecting operation parameters of the harmonic reducer at regular intervals (such as 1s), and collecting operation and monitoring parameters of a variable frequency motor, a sensor and the like;
step 2: taking the harmonic reducer operating parameters obtained in the step 1 as an original data set, predicting a power signal at the next moment, and selecting monitoring data of a certain time step length to obtain a sample data set, for example: intercepting 10000 data points of normal operation of the harmonic reducer as an original data set, and carrying out normalization processing on the original data set;
and step 3: dividing the normalized data into a training set, a verification set and a test set according to a certain proportion; the training set and the verification set train the mixed deep neural network, and the test set is used for checking the effectiveness of the mixed deep neural network;
more specifically, the data set is constructed using a sliding window method, with a time domain window length of 20 being chosen. Dividing a training set and a test set according to a ratio of 7:3, and dividing an initial training set into a training set and a verification set according to a ratio of 9:1 during training;
and 4, step 4: training the hybrid deep neural network by using training set data and obtaining corresponding weight and bias, predicting a power signal of a verification set by using the obtained weight and bias, calculating a prediction error of the verification set data, and adjusting hyper-parameters of the trained hybrid deep neural network when the prediction error does not meet preset requirements, such as: a learning rate; storing the currently adjusted trained mixed deep neural network until the prediction error meets the preset requirement; the structure diagram of the hybrid deep neural network model is shown in fig. 2, and the specific training steps are as follows: the specific training flow of the hybrid deep neural network model is shown in fig. 3;
step (1): initializing parameters of a CNN layer, a BilSTM layer and a full connection layer, optimizing by using a random gradient descent Adam algorithm, and setting an early termination training condition Q and a maximum training period number of 100;
step (2): filtering noise of input data and extracting spatial coupling characteristics by using a CNN branch, extracting bidirectional time sequence dependence of the input data by using a BilSTM branch, finally fusing data characteristics extracted by the two branches, and completing prediction of the power of a harmonic reducer at the next moment;
and (3): calculating a gradient and updating a hybrid depth neural network weight parameter using an Adam optimizer; if the training condition for early termination is not reached or the number of training sessions is less than 100, the step (2) is circulated, otherwise, the training is terminated;
and (4): adjusting the hyper-parameters of the hybrid deep neural network model by using the verification set, carrying out preliminary evaluation on the prediction performance of the model, and finally storing the hybrid deep neural network meeting the preset conditions;
and 5: and inputting the data of the corresponding test set into the trained mixed deep neural network, and predicting the power signal at the next moment.
Fig. 4 and 5 are schematic diagrams of power prediction results on a test set in a harmonic reducer operating parameter data set based on a quadruple load experiment condition according to an embodiment of the invention. Meanwhile, the mixed deep neural network model has an MSE of 0.000604, an MAE of 0.018679 and an MAPE of 3.6% on the test set.
Those skilled in the art will appreciate that, in addition to implementing the systems, apparatus, and various modules thereof provided by the present invention in purely computer readable program code, the same procedures can be implemented entirely by logically programming method steps such that the systems, apparatus, and various modules thereof are provided in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system, the device and the modules thereof provided by the present invention can be considered as a hardware component, and the modules included in the system, the device and the modules thereof for implementing various programs can also be considered as structures in the hardware component; modules for performing various functions may also be considered to be both software programs for performing the methods and structures within hardware components.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (10)
1. A harmonic reducer power real-time prediction method based on a hybrid deep neural network is characterized by comprising the following steps:
step S1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
step S2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
step S3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
step S4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
step S5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
step S6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
2. The hybrid deep neural network-based harmonic reducer power real-time prediction method according to claim 1, wherein the acquiring of the operation and monitoring parameters of the harmonic reducer in a normal operation state comprises: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
3. The method for predicting the power of the harmonic reducer based on the hybrid deep neural network in real time according to claim 1, wherein the preprocessing the data set in the step S2 comprises: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
4. The method for real-time prediction of harmonic reducer power based on hybrid deep neural network as claimed in claim 1, wherein the hybrid deep neural network comprises: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
5. The method for predicting the power of the harmonic reducer based on the hybrid deep neural network in real time according to claim 4, wherein the step S4 comprises: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
6. A harmonic reducer power real-time prediction system based on a hybrid deep neural network is characterized by comprising the following components:
module M1: building an accelerated life test platform of the harmonic reducer, and collecting operation and monitoring parameters of the harmonic reducer in a normal operation state;
module M2: constructing a data set based on the collected operation and monitoring parameters of the harmonic reducer, and preprocessing the data set to obtain a preprocessed data set;
module M3: dividing the preprocessed data set into a training set, a verification set and a test set according to a preset proportion;
module M4: training the mixed deep neural network by using a training set and obtaining corresponding weight and bias to obtain the trained mixed deep neural network;
module M5: predicting a power signal of a verification set by using the trained mixed deep neural network, calculating a prediction error of data of the verification set, adjusting a hyper-parameter of the trained mixed deep neural network when the prediction error does not meet a preset requirement until the prediction error meets the preset requirement, and storing the currently adjusted trained mixed deep neural network;
module M6: and inputting the data of the corresponding test set into the trained mixed deep neural network after adjustment, and predicting the power signal at the next moment.
7. The hybrid deep neural network-based harmonic reducer power real-time prediction system of claim 6, wherein the collecting operating and monitoring parameters of the harmonic reducer in normal operating conditions comprises: the operation and monitoring parameters of the variable frequency motor, the magnetic powder brake and the sensor;
the operating and monitoring parameters include: inputting torque rotating speed, temperature, acceleration in three directions of an X axis, a Y axis and a Z axis, and outputting torque rotating speed and power signals.
8. The hybrid deep neural network-based harmonic reducer power real-time prediction system of claim 6, wherein the preprocessing of the data set in the module M2 comprises: and carrying out normalization processing on the data set by using a maximum and minimum normalization method to obtain a preprocessed data set.
9. The hybrid deep neural network-based harmonic reducer power real-time prediction system of claim 6, wherein the hybrid deep neural network comprises: extracting spatial features and bidirectional time sequence dependence features of the state monitoring data respectively by using the convolutional neural network branches and the bidirectional long and short term memory neural network branches, connecting the outputs of the convolutional neural network and the bidirectional long and short term memory neural network, and obtaining an output result of the hybrid deep neural network through a full connection layer network of a preset layer;
the convolutional neural network branch comprises a preset layer CNN convolutional layer, a preset layer maximum pooling operation layer and a preset layer flattening operation layer; the convolutional neural network is used for filtering noise of input data and extracting spatial characteristics of monitoring data;
the bidirectional long-short term memory neural network branch comprises a preset layer BilSTM layer stack, the BilSTM layer avoids gradient loss and gradient explosion when training the mixed deep neural network, and simultaneously extracts the time sequence characteristics of monitoring data.
10. The hybrid deep neural network-based harmonic reducer power real-time prediction system of claim 9, wherein the module M4 includes: initializing parameters of a CNN convolutional layer, a BilSTM layer and a full-connection layer in the hybrid deep neural network, and optimizing weight parameters and bias of the hybrid deep neural network by using a random gradient descent Adam algorithm until a loss function is minimum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110672801.2A CN113297704A (en) | 2021-06-17 | 2021-06-17 | Harmonic reducer power real-time prediction method and system based on hybrid deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110672801.2A CN113297704A (en) | 2021-06-17 | 2021-06-17 | Harmonic reducer power real-time prediction method and system based on hybrid deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113297704A true CN113297704A (en) | 2021-08-24 |
Family
ID=77328573
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110672801.2A Pending CN113297704A (en) | 2021-06-17 | 2021-06-17 | Harmonic reducer power real-time prediction method and system based on hybrid deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113297704A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113752300A (en) * | 2021-09-22 | 2021-12-07 | 重庆工商大学 | Industrial robot energy consumption prediction method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6119112A (en) * | 1997-11-19 | 2000-09-12 | International Business Machines Corporation | Optimum cessation of training in neural networks |
| US20030203717A1 (en) * | 1998-04-27 | 2003-10-30 | Chuprun Jeffery Scott | Satellite based data transfer and delivery system |
| CN102184453A (en) * | 2011-05-16 | 2011-09-14 | 上海电气集团股份有限公司 | Wind power combination predicting method based on fuzzy neural network and support vector machine |
| CN104554251A (en) * | 2014-12-09 | 2015-04-29 | 河南理工大学 | Hybrid car energy-saving prediction control method based on road gradient information |
| CN105701348A (en) * | 2016-01-12 | 2016-06-22 | 芜湖赛宝信息产业技术研究院有限公司 | Method for predicting reliable life of harmonic reducer |
| CN112819236A (en) * | 2021-02-07 | 2021-05-18 | 南京邮电大学 | Low-orbit satellite Internet of things active terminal prediction method based on neural network |
-
2021
- 2021-06-17 CN CN202110672801.2A patent/CN113297704A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6119112A (en) * | 1997-11-19 | 2000-09-12 | International Business Machines Corporation | Optimum cessation of training in neural networks |
| US20030203717A1 (en) * | 1998-04-27 | 2003-10-30 | Chuprun Jeffery Scott | Satellite based data transfer and delivery system |
| CN102184453A (en) * | 2011-05-16 | 2011-09-14 | 上海电气集团股份有限公司 | Wind power combination predicting method based on fuzzy neural network and support vector machine |
| CN104554251A (en) * | 2014-12-09 | 2015-04-29 | 河南理工大学 | Hybrid car energy-saving prediction control method based on road gradient information |
| CN105701348A (en) * | 2016-01-12 | 2016-06-22 | 芜湖赛宝信息产业技术研究院有限公司 | Method for predicting reliable life of harmonic reducer |
| CN112819236A (en) * | 2021-02-07 | 2021-05-18 | 南京邮电大学 | Low-orbit satellite Internet of things active terminal prediction method based on neural network |
Non-Patent Citations (3)
| Title |
|---|
| CHENGYING ZHAO 等: "A Double-Channel Hybrid Deep Neural Network Based on CNN and BiLSTM for Remaining Useful Life Prediction", 《SENSORS》 * |
| RENFANG HUANG 等: "Energy performance prediction of the centrifugal pumps by using a hybrid neural network", 《ENERGY》 * |
| YIJUN YOO 等: "Accelerated Life Calculation for Harmonic reduction of Robot Arm Using Weibull Distribution", 《2012 9TH INTERNATIONAL CONFERENCE ON UBIQUITOUS ROBOTS AND AMBIENT INTELLIGENCE (URAI)》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113752300A (en) * | 2021-09-22 | 2021-12-07 | 重庆工商大学 | Industrial robot energy consumption prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102520697B (en) | Onsite information preprocessing method of remote cooperative diagnosis | |
| CN107730145B (en) | Voltage sag economic loss assessment method | |
| US20050273296A1 (en) | Neural network model for electric submersible pump system | |
| CN114429153A (en) | Lifetime learning-based gearbox increment fault diagnosis method and system | |
| CN109472097B (en) | Fault diagnosis method for online monitoring equipment of power transmission line | |
| CN115562225A (en) | Industrial robot operation and maintenance management method and device, computer equipment and storage medium | |
| CN113343591A (en) | Product key part life end-to-end prediction method based on self-attention network | |
| CN115204491A (en) | Production line working condition prediction method and system based on digital twinning and LSTM | |
| CN113297704A (en) | Harmonic reducer power real-time prediction method and system based on hybrid deep neural network | |
| CN112257263A (en) | Equipment residual life prediction system based on self-attention mechanism | |
| CN111581831B (en) | Failure-related multi-state system reliability assessment method | |
| Wang et al. | Fault diagnosis and predictive maintenance for hydraulic system based on digital twin model | |
| Ge et al. | Remaining useful life prediction of machinery based on KS distance and LSTM neural network | |
| Morsalin et al. | Induction motor inter-turn fault detection using heuristic noninvasive approach by artificial neural network with Levenberg Marquardt algorithm | |
| CN114583767B (en) | Data-driven wind power plant frequency modulation response characteristic modeling method and system | |
| Junhong et al. | Software reliability nonlinear modeling and its fuzzy evaluation | |
| Zhou et al. | Robot Remote Monitoring and Fault Diagnosis Based on Industrial Internet of Things | |
| CN112926227A (en) | System reliability evaluation method and system based on subcomponent reliability function | |
| Li et al. | Failure Warning of Harmonic Reducer Based on Power Prediction | |
| Zhou et al. | Bearing life prediction method based on parallel multichannel recurrent convolutional neural network | |
| Dewantoro et al. | Optimized PID-Like Neural Network Controller for Single-Objective Systems | |
| CN113761792B (en) | Neural network-based six-pulse rectifier fault diagnosis method | |
| Bahiuddin et al. | A Systematic Approach to Estimate Non-Linear System Parameters using Particle Swarm Optimization and Bond Graph Methods | |
| Lee et al. | A Study on the predictive maintenance system for CNC Facilities based on intelligent cutting tool load prediction | |
| US20220299988A1 (en) | System, apparatus and method of condition based management of one or more electro-mechanical systems |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210824 |