CN109782167A - Quality inspection device and method for gear reducer motor based on convolutional network depth model - Google Patents
Quality inspection device and method for gear reducer motor based on convolutional network depth model Download PDFInfo
- Publication number
- CN109782167A CN109782167A CN201811633174.6A CN201811633174A CN109782167A CN 109782167 A CN109782167 A CN 109782167A CN 201811633174 A CN201811633174 A CN 201811633174A CN 109782167 A CN109782167 A CN 109782167A
- Authority
- CN
- China
- Prior art keywords
- motor
- model
- electric motor
- convolutional network
- toothed gearing
- 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
- 238000000034 method Methods 0.000 title claims abstract description 22
- 238000007689 inspection Methods 0.000 title abstract 3
- 239000003638 chemical reducing agent Substances 0.000 title 1
- 230000001133 acceleration Effects 0.000 claims abstract description 21
- 230000001105 regulatory effect Effects 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 17
- 238000009432 framing Methods 0.000 claims description 8
- 239000002184 metal Substances 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 6
- 229910000831 Steel Inorganic materials 0.000 claims description 5
- 239000010959 steel Substances 0.000 claims description 5
- 241000208340 Araliaceae Species 0.000 claims description 4
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 4
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 4
- 235000008434 ginseng Nutrition 0.000 claims description 4
- 239000003292 glue Substances 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 9
- 230000009467 reduction Effects 0.000 abstract description 4
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 239000000306 component Substances 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 206010016256 fatigue Diseases 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 229910001220 stainless steel Inorganic materials 0.000 description 1
- 239000010935 stainless steel Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
- 230000036642 wellbeing Effects 0.000 description 1
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
本发明公开了一种基于卷积网络深度模型的齿轮减速电机品检装置及方法，所述装置包括：直流稳压电源，用于为给待检测微型直流齿轮减速电机供电；恒流适配器，用于为加速度传感器供电，以及将加速度传感器采集的电机振动信号放大后输送至数据采集器；数据采集器，用于将放大后的电机振动信号生成数字信号输送中PC机；加速度传感器，用于采集待检测微型直流齿轮减速电机空转情况下的电机振动信号；PC机，用于根据所述数字信号判断待检测微型直流齿轮减速电机的优劣。本发明解决了目前在电机品检领域广泛使用的人工检测法所带来的巨大劳动成本和疲惫检测等问题，在保证精确率的同时，提高对微型直流齿轮减速电机的品质鉴定的效率。
The invention discloses a gear deceleration motor quality inspection device and method based on a convolution network depth model. The device comprises: a DC regulated power supply for supplying power to a micro DC gear deceleration motor to be detected; a constant current adapter for It is used to power the acceleration sensor and amplify the motor vibration signal collected by the acceleration sensor and then send it to the data collector; the data collector is used to generate a digital signal from the amplified motor vibration signal and transmit it to the PC; the acceleration sensor is used to collect The motor vibration signal when the micro DC gear reduction motor to be detected is idling; the PC is used to judge the pros and cons of the micro DC gear reduction motor to be detected according to the digital signal. The invention solves the problems of huge labor cost and fatigue detection caused by the manual detection method widely used in the field of motor quality inspection at present, and improves the quality identification efficiency of the micro DC gear reduction motor while ensuring the accuracy.
Description
Technical field
The present invention relates to technology for mechanical fault diagnosis more particularly to a kind of minisize dcs based on convolutional network depth model
Toothed gearing electric motor product examine device and method.
Background technique
With the mankind science and technology it is increasingly developed, automation will become 21 century production development one of theme.And conduct
The motor that mechanical energy can be converted electrical energy into always is the indispensable core component of each field automated system.Wherein,
In low speed, the occasion of big torque, toothed gearing electric motor is always most economical, practical preferred option.
Gear reduction box is exactly installed on the output shaft of motor by so-called toothed gearing electric motor, passes through subtracting for gear
Speed by output revolving speed by being reduced to low speed at a high speed, while improving output torque.Due to the characteristic, such micromotor product is extensive
For automatic production line, in the precision instruments such as Medical Devices and intelligent industrial, reading intelligent agriculture, smart home, intelligence machine
In the output of the relevant devices intelligent power such as people.Under the market competitive pressure of such fierceness of such product, how extensive
The quality for guaranteeing product while production, will can become toothed gearing electric motor enterprise create the weight of considerable economic well-being of workers and staff
Want one of problem.
At present to the Quality Detection of minisize dc toothed gearing electric motor, in addition to several rigid index revolving speeds, torque, temperature rise
Deng outside, also noise and gear quality are identified.And at home in miniature gears decelerating motor factory, part detection is general
It is to perceive the vibration of empty load of motor by both hands and listen to empty load of motor with ear all over being identified by manual type
Noise carries out the superiority and inferiority of comprehensive judgement product.This backward ineffective technique not only greatly increases the labour cost in production, and
And since this repeated labor can make worker error in judgement tired out occur, market is marched toward so as to cause substandard products, to the letter of enterprise
The loss that can not be retrieved is carried out in reputation and subsequent economy-zone.
Summary of the invention
In view of the above technical problems, the present invention is directed to depth learning technology is applied to minisize dc toothed gearing electric motor
In Quality Detection, the precision and efficiency of motor product examine can be significantly improved using the method for deep learning, reduces the people of enterprise
The problems such as power cost and precision efficiency.
The present invention adopts the following technical scheme that realization:
A kind of toothed gearing electric motor product examine device based on convolutional network depth model, comprising:
D.C. regulated power supply, for being to power to minisize dc toothed gearing electric motor to be detected；
Constant current adapter is believed for powering for acceleration transducer, and by the motor oscillating of acceleration transducer acquisition
Number amplification after be delivered to data collector；
Data collector, for amplified motor oscillating signal to be generated PC machine in digital signal conveying；
Acceleration transducer, for acquiring the motor oscillating letter in the case of minisize dc toothed gearing electric motor to be detected dallies
Number；
PC machine, for judging the superiority and inferiority of minisize dc toothed gearing electric motor to be detected according to the digital signal.
Further, the minisize dc toothed gearing electric motor to be detected passes through with a thickness of 0.1-0.25mm rectangular metal
Steel disc is contacted with acceleration transducer surface rigidity.
Further, acceleration transducer measurement surface realizes that rigidity connects by metal glue with metal steel disc
It connects.
Further, the D.C. regulated power supply uses digital control type linear direct current regulated power supply；The constant current adapter is adopted
With single-channel constant current adapter；The data collector uses usb multifunctional data acquisition card.
A kind of toothed gearing electric motor product examine method based on convolutional network depth model should using such as described product examine device
Method comprising steps of
1) vibration signal for extracting minisize dc toothed gearing electric motor to be measured switchs to digital voltage signal and is passed in PC machine；
2) PC machine obtains the vibration signal of motor, and pretreatment obtains the corresponding triple channel time-frequency figure of vibration signal；
3) the triple channel time-frequency figure is inputted into the convolutional network depth point that selected corresponding motor model training finishes
In class model, the classification information of motor superiority and inferiority is identified, to mitigate the burden of artificial product examine.
Further, the pretreatment obtains the corresponding triple channel time-frequency figure of vibration signal and specifically includes: obtaining five seconds electricity
Then machine time-domain signal carries out adding window, framing, Fast Fourier Transform (FFT) and image standardized processing to signal and obtains triple channel
Five seconds time-frequency figures of signal.
Further, the training of the convolutional network depth sorting model specifically includes:
1) Signals Data Base of minisize dc toothed gearing electric motor is established, i.e. collection specific model superiority and inferiority motor is each hundreds of
Platform, processing obtain five seconds time-frequency data-signals of each motor, and divide training set and test set and verifying according to a certain percentage
Collection；
2) adding window framing carried out to five seconds time-frequency data-signals, carried out at Fast Fourier Transform (FFT) and image standardization
Reason obtains five seconds time-frequency figures of signal of triple channel；
3) convolutional network depth sorting model is then built by tensorflow deep learning library, first initialization model is super
Parameter, including learning rate, batch size, CNN core size and quantity, network frame position and quantity, then in convolutional layer and Quan Lian
Layer addition batch normalized is connect, dropout layers are added after full articulamentum, and L2 regularization is done to weight, is put into last
During LOSS is calculated.This part is the pace of learning and model generalization ability in order to improve model, and model is allowed to possess test set mistake
The characteristics of accidentally rate is low, energy real-time detection motor quality.
4) data of training set are inputted into training in the convolutional network Deep model of initialization, 20 periods or
When the opposite variation of LOSS value for network occur is less than threshold value, the ginseng such as weight of convolutional network Deep model is adjusted by BP algorithm
Number；
5) data of verifying collection are inputted into verifying accuracy rate in the trained model of step 6, it is big if there is accuracy rate
In threshold value, then prototype network parameter is kept, step 3) is otherwise returned to, adjusts the network frame and hyper parameter of model.
Further, the first layer of the convolutional network deep layer disaggregated model uses convolution kernel for the one-dimensional convolution net of 7*1
Network, it is subsequent using the full connection layer network of one layer of maxpool, two layers of two-dimensional convolution network and two layers, divided by softmax
Class.
Further, the model hyper parameter specifically:
The feature of input is the triple channel time-frequency figure of 3*129*92, and the batchsize of network is 40, and learning rate is that 1e-4 is right
1e-5 is changed to the 20th period afterwards.
Further, the classification information for identifying motor superiority and inferiority specifically includes: by dividing to five seconds motor signals
And judge respectively, it is judged as excellent motor if the excellent result of appearance is more；If obtained softmax result is more than certain there are ratio
Threshold value is then furnished position again and is measured to motor.
Compared with prior art, the present invention can accurately extract the vibration signal of minisize dc toothed gearing electric motor, in conjunction with building
The vertical depth model library based on convolutional network judges the superiority and inferiority situation of minisize dc toothed gearing electric motor, improves to miniature straight
The efficiency of the quality identification of toothed gearing electric motor is flowed, is solved huge brought by the widely used artificial detection method in the field at present
The problems such as big labour cost and detection tired out, and then reduce the workload manually judged and improve electric machines test precision, to mention
High motor production efficiency.
Detailed description of the invention
Fig. 1 is the product examine system structure diagram of the toothed gearing electric motor based on convolutional network.
Fig. 2 is the identification process figure of the product examine system of toothed gearing electric motor.
Fig. 3 is the model buildings training flow chart of the minisize dc toothed gearing electric motor based on convolution deep layer network.
Fig. 4 is the frame and partial parameters of convolutional network Deep model.
In figure: 1- D.C. regulated power supply；2- minisize dc toothed gearing electric motor to be detected；3- metal steel disc；4- acceleration
Sensor；5- constant current adapter；6- data collector；7-PC machine.
Specific embodiment
The present invention is described further in the following with reference to the drawings and specific embodiments.
As shown in Figure 1, a kind of toothed gearing electric motor product examine device based on convolutional network depth model, comprising:
D.C. regulated power supply 1, for power to minisize dc toothed gearing electric motor 2 to be detected, the present embodiment to be used
MOTECH digital control type linear direct current regulated power supply LPS-305；
Constant current adapter 5, for being the power supply of acceleration transducer 4, and the motor oscillating that acceleration transducer 4 is acquired
Data collector 6 is delivered to after signal amplification, the present embodiment uses CT5201 single-channel constant current adapter；
Data collector 6, for amplified motor oscillating signal to be generated PC machine 7 in digital signal conveying, this implementation
Example uses MCC1608G usb multifunctional data acquisition card DAQ；Amplified motor oscillating signal is transported to MCC1608G usb
Behind the simulation input port of multifunctional data acquisition card DAQ, analog signal can be converted to digital signal and is passed to by data collector
Into PC machine 7；
Acceleration transducer 3, for acquiring the motor oscillating in the case of minisize dc toothed gearing electric motor 2 to be detected dallies
Signal, the present embodiment use CT1050LC acceleration transducer, and the minisize dc toothed gearing electric motor to be detected passes through thickness
The rectangle stainless steel substrates and 3 surface of acceleration transducer that degree is 0.25mm pass through metal glue realization rigid contact；
PC machine 7, for judging the superiority and inferiority of minisize dc toothed gearing electric motor 2 to be detected according to the digital signal.
As shown in Fig. 2, a kind of toothed gearing electric motor product examine method based on convolutional network depth model, using such as described product
Checking device, the method comprising the steps of:
1) five seconds motor oscillating signals for extracting minisize dc toothed gearing electric motor to be measured, are then split into five parts of one second electricity
Machine vibration signal switchs to digital voltage signal and is passed in PC machine；
2) PC machine obtains the vibration signal of motor, then to five parts of one second motor oscillating signals carry out adding windows, framing,
Fast Fourier Transform (FFT) and image standardized processing obtain five seconds time-frequency figures of signal of triple channel.
3) the triple channel time-frequency figure is inputted into the convolutional network depth point that selected corresponding motor model training finishes
In class model, obtain five parts of softmax as a result, the small Mr. Yu's threshold value of ratio if there is some softmax, illustrate model without
Method accurately separates motor superiority and inferiority, needs to return to step 1 and is re-measured；Then judge five parts of softmax as a result, such as
Fruit majority result belongs to certain one kind, then judges that the motor quality for such, identifies the classification information of motor superiority and inferiority, to mitigate
The burden of artificial product examine.
Fig. 3 is the training flow chart of the convolutional network depth sorting model, comprising steps of
Step 1: according to the artificial judgment method of vestibule, each 100, superiority and inferiority motor are collected, and is marked.
Step 2: acquiring 5s motor oscillating signal to every sample motor by the product examine device, and each 20 groups.According to 6:
2:2 ratio cut partition training set, test set and verifying collection, and 5s motor signal is divided into 5 parts of 1s motor signals in order.
Step 3: since motor signal belongs to non-stationary signal, Fourier transformation can just be done by needing to do adding window sub-frame processing,
Wherein window function selection is hamming window:
Weaken to offset adding window bring both ends, so needing to need be overlapped, frame when framing, between frame and frame
A length of 256 sampled point, it is 128 sampled points that frame, which moves,.
Step 4: Fast Fourier Transform (FFT) is carried out to the motor signal after adding window framing, obtains corresponding triple channel motor
Time-frequency figure.Then 2 standardization of image is done to every picture:
Wherein x indicates that image array, μ are that the identical matrix of size, element are the mean value of image therewith.σ indicates that standard variance, N indicate the pixel number of image x.Image standardization
It is by going mean value to realize the processing of centralization to be easier that model is allowed to obtain extensive effect in this way by data.
Step 5: convolutional network Deep model, trained model framework are built by tensorflow deep learning library
As shown in Fig. 3, first initialization model hyper parameter, including learning rate at present, batch size, CNN core size and quantity, network frame
Position and quantity etc..Then in convolutional layer and full articulamentum addition batch normalized, dropout is added after full articulamentum
Layer, and L2 regularization is done to weight, it is put into last LOSS calculating.This part is the pace of learning and mould in order to improve model
Type generalization ability.
Step 6: the data of training set are inputted into training in the initialization model of step 5,20 periods or
When there is the LOSS value small Mr. Yu's threshold value of opposite variation of network, passes through BP algorithm and adjust the ginseng such as weight of convolutional network Deep model
Number.
Step 7: the data of verifying collection are inputted into accuracy rate is verified in the trained model of step 6, if there is standard
True rate is greater than threshold value, then keeps prototype network parameter, otherwise return to step 5, adjusts the network frame of model and surpasses ginseng
Number.
As shown in figure 4, the first layer of the embodiment of the present invention convolutional network deep layer disaggregated model uses convolution kernel for 7*1's
One-dimensional convolutional network, it is subsequent using the full connection layer network of one layer of maxpool, two layers of two-dimensional convolution network and two layers, pass through
Softmax classifies.The feature of input is the triple channel time-frequency figure of 3*129*92, and the batchsize of network is 40, study
Rate is 1e-4, is then changed to 1e-5 the 20th period.
In conclusion the embodiment of the present invention mainly includes hardware components and software section, in which:
Hardware components: mainly including acceleration transducer, D.C. regulated power supply, constant current adapter, data collector, PC
Machine.By D.C. regulated power supply give minisize dc toothed gearing electric motor power supply, under no-load condition with acceleration transducer surface
Rigid contact；Powered by constant current adapter and amplify motor vibration signal and data collector generate PC machine number letter
Number, the algorithm designed using the present invention can accurately extract the vibration signal of minisize dc toothed gearing electric motor, in conjunction with foundation
Depth model library based on convolutional network judges the superiority and inferiority situation of minisize dc toothed gearing electric motor, and then reduces artificial tasting
Workload and improve electric machines test precision, to improve motor production efficiency.
Software section: following a few class contents are related generally to:
1, the frequency-region signal of vibration signal: adding window framing is carried out to signal using window function, and each frame signal is carried out fast
Fast Fourier transformation and five seconds signal time-frequency figures for generating corresponding triple channel；
2, the standardization of image: by every image by go mean value realize centralization processing, according to convex optimum theory with
Data probability distributions relevant knowledge, data center meet data distribution rule, it is easier to obtain extensive effect after training；
3, convolutional network is built and is learnt: according to the suitable deep layer convolutional Neural of the signal characteristic design parameter of acquisition point
The hyper parameter of convolutional network and the position and quantity of maxpool and full articulamentum is arranged in class network.Meeting precision while protecting
The generalization ability of model of a syndrome；
4, the optimization of convolutional network: addition batch standardization, the technologies such as L2 regularization and dropout improve the study speed of model
Rate and generalization ability；
5, the identification of telecommunication signal: identifying by five seconds motor signals of segmentation and respectively to improve the discrimination of network, with
And the degree of reliability of classification is measured by judging the ratio of softmax.
The embodiment of the present invention is directed to the problems such as Quality Detection efficiency and precision of minisize dc toothed gearing electric motor, will be deep
The method of degree study incorporates wherein.Feature is carried out by vibration signal of the improved convolutional neural networks to toothed gearing electric motor to mention
It takes and model learning, realizes the Quality Detection task of minisize dc toothed gearing electric motor.While guaranteeing accurate rate, improve
To the efficiency of the quality identification of minisize dc toothed gearing electric motor.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
In this, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention patent
Technical solution and its patent of invention design are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.
Claims (10)
1. a kind of toothed gearing electric motor product examine device based on convolutional network depth model characterized by comprising
D.C. regulated power supply, for being to power to minisize dc toothed gearing electric motor to be detected；
Constant current adapter is put for powering for acceleration transducer, and by the motor oscillating signal of acceleration transducer acquisition
Data collector is delivered to after big；
Data collector, for amplified motor oscillating signal to be generated PC machine in digital signal conveying；
Acceleration transducer, for acquiring the motor oscillating signal in the case of minisize dc toothed gearing electric motor to be detected dallies；
PC machine, for judging the superiority and inferiority of minisize dc toothed gearing electric motor to be detected according to the digital signal.
2. the toothed gearing electric motor product examine device according to claim 1 based on convolutional network depth model, feature exist
In the minisize dc toothed gearing electric motor to be detected with a thickness of 0.1-0.25mm rectangular metal steel disc and acceleration by passing
Sensor surfaces rigid contact.
3. the toothed gearing electric motor product examine device according to claim 1 based on convolutional network depth model, feature exist
In the acceleration transducer measurement surface and metal steel disc are realized by metal glue and be rigidly connected.
4. the toothed gearing electric motor product examine device according to claim 1 based on convolutional network depth model, feature exist
In the D.C. regulated power supply uses digital control type linear direct current regulated power supply；The constant current adapter is suitable using single-channel constant current
Orchestration；The data collector uses usb multifunctional data acquisition card.
5. a kind of toothed gearing electric motor product examine method based on convolutional network depth model, which is characterized in that wanted using such as right
Product examine device described in asking any one of 1 to 4, the method comprising the steps of:
The vibration signal for extracting minisize dc toothed gearing electric motor to be measured switchs to digital voltage signal and is passed in PC machine；
The PC machine obtains the vibration signal of motor, and pretreatment obtains the corresponding triple channel time-frequency figure of vibration signal；
The triple channel time-frequency figure is inputted into the convolutional network depth sorting model that selected corresponding motor model training finishes
In, identify the classification information of motor superiority and inferiority.
6. the toothed gearing electric motor product examine method according to claim 5 based on convolutional network depth model, feature exist
In the pretreatment obtains the corresponding triple channel time-frequency figure of vibration signal and specifically includes: obtaining five seconds motor time-domain signals, then
Five seconds time-frequencies of signal that adding window, framing, Fast Fourier Transform (FFT) and image standardized processing obtain triple channel are carried out to signal
Figure.
7. the toothed gearing electric motor product examine method according to claim 5 based on convolutional network depth model, feature exist
In the training of the convolutional network depth sorting model specifically includes:
1) Signals Data Base of minisize dc toothed gearing electric motor is established, i.e. collection each hundreds of specific model superiority and inferiority motor, place
Reason obtains five seconds time-frequency data-signals of each motor, and divides training set and test set and verifying collection according to a certain percentage；
2) adding window framing, progress Fast Fourier Transform (FFT) and image standardized processing is carried out to five seconds time-frequency data-signals to obtain
Obtain five seconds time-frequency figures of signal of triple channel；
3) convolutional network depth sorting model, the super ginseng of first initialization model are then built by tensorflow deep learning library
Number, including learning rate, batch size, CNN core size and quantity, network frame position and quantity, then in convolutional layer and full connection
Layer addition batch normalized, adds dropout layers, and do L2 regularization to weight after full articulamentum, is put into last LOSS
In calculating；
4) data of training set are inputted into training in the convolutional network Deep model of initialization, in 20 periods or appearance
When the opposite variation of the LOSS value of network is less than threshold value, the parameters such as the weight of convolutional network Deep model are adjusted by BP algorithm；
5) data of verifying collection are inputted into accuracy rate is verified in the trained model of step 6, is greater than threshold if there is accuracy rate
Value, then keep prototype network parameter, otherwise return to step 3), adjust the network frame and hyper parameter of model.
8. the toothed gearing electric motor product examine method according to claim 5 based on convolutional network depth model, feature exist
In the first layer of the convolutional network deep layer disaggregated model uses convolution kernel for the one-dimensional convolutional network of 7*1, subsequent to use one layer
The full connection layer network of maxpool, two layers of two-dimensional convolution network and two layers, is classified by softmax.
9. the toothed gearing electric motor product examine method according to claim 7 based on convolutional network depth model, feature exist
In the model hyper parameter specifically:
The feature of input is the triple channel time-frequency figure of 3*129*92, and the batchsize of network is 40, and learning rate is then 1e-4 exists
20th period is changed to 1e-5.
10. the toothed gearing electric motor product examine method according to claim 5 based on convolutional network depth model, feature exist
In the classification information for identifying motor superiority and inferiority specifically includes: by dividing to five seconds motor signals and judging respectively, if going out
Existing excellent result is more, is judged as excellent motor；If obtained softmax result is more than certain threshold value there are ratio, to motor weight
New ornaments position simultaneously measures.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811633174.6A CN109782167B (en) | 2018-12-29 | 2018-12-29 | Gear reduction motor quality inspection device and method based on convolutional network depth model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811633174.6A CN109782167B (en) | 2018-12-29 | 2018-12-29 | Gear reduction motor quality inspection device and method based on convolutional network depth model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109782167A true CN109782167A (en) | 2019-05-21 |
| CN109782167B CN109782167B (en) | 2024-05-10 |
Family
ID=66498890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811633174.6A Active CN109782167B (en) | 2018-12-29 | 2018-12-29 | Gear reduction motor quality inspection device and method based on convolutional network depth model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109782167B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110109015A (en) * | 2019-05-31 | 2019-08-09 | 中南大学 | A kind of asynchronous motor Fault monitoring and diagnosis method based on deep learning |
| CN110705456A (en) * | 2019-09-29 | 2020-01-17 | 华南理工大学 | Micro motor abnormity detection method based on transfer learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107421741A (en) * | 2017-08-25 | 2017-12-01 | 南京信息工程大学 | A kind of Fault Diagnosis of Roller Bearings based on convolutional neural networks |
| CN209280134U (en) * | 2018-12-29 | 2019-08-20 | 华南理工大学 | A kind of minisize dc toothed gearing electric motor product examine device based on acceleration transducer |
-
2018
- 2018-12-29 CN CN201811633174.6A patent/CN109782167B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107421741A (en) * | 2017-08-25 | 2017-12-01 | 南京信息工程大学 | A kind of Fault Diagnosis of Roller Bearings based on convolutional neural networks |
| CN209280134U (en) * | 2018-12-29 | 2019-08-20 | 华南理工大学 | A kind of minisize dc toothed gearing electric motor product examine device based on acceleration transducer |
Non-Patent Citations (2)
| Title |
|---|
| TURKER INCE 等: "Real-Time Motor Fault Detection by 1-D Convolutional Neural Networks", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》, vol. 63, no. 11, pages 7067 - 7075, XP011625199, DOI: 10.1109/TIE.2016.2582729 * |
| 罗炳聪等: "小型电机生产过程在线品质检测方法研究", 《测控技术》, vol. 34, no. 12, pages 34 - 37 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110109015A (en) * | 2019-05-31 | 2019-08-09 | 中南大学 | A kind of asynchronous motor Fault monitoring and diagnosis method based on deep learning |
| CN110705456A (en) * | 2019-09-29 | 2020-01-17 | 华南理工大学 | Micro motor abnormity detection method based on transfer learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109782167B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN108590982B (en) | Abnormal data processing method for limited power operation of wind turbine generator | |
| CN104697767B (en) | Rotor system fault diagnosis method and device based on vibration analysis | |
| CN111830408A (en) | Motor fault diagnosis system and method based on edge calculation and deep learning | |
| CN110333077B (en) | A Fault Diagnosis Method of Rolling Bearing Based on Approximate Support Vector | |
| CN109323754A (en) | A Method for Diagnosing and Detecting Polygon Faults of Train Wheels | |
| CN110991055B (en) | Residual life prediction system for rotary mechanical products | |
| CN101710235A (en) | Method for automatically identifying and monitoring on-line machined workpieces of numerical control machine tool | |
| Song et al. | On the accuracy of fault diagnosis for rolling element bearings using improved DFA and multi-sensor data fusion method | |
| CN113537160B (en) | Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium | |
| CN111878322B (en) | Wind Turbine Device | |
| CN109782167A (en) | Quality inspection device and method for gear reducer motor based on convolutional network depth model | |
| CN117066337B (en) | Stamping control method and system for motor stator and rotor machining | |
| CN106271881A (en) | A kind of Condition Monitoring of Tool Breakage method based on SAEs and K means | |
| CN109254577A (en) | A kind of intelligence manufacture procedure fault classification method and device based on deep learning | |
| CN109947047A (en) | A method for diagnosing unbalance faults of electric spindle | |
| CN109605128B (en) | An online detection method of milling chatter based on power spectrum entropy difference | |
| CN115954017A (en) | HHT-based engine small sample sound abnormal fault identification method and system | |
| Zhang et al. | Complementary ensemble adaptive local iterative filtering and its application to rolling bearing fault diagnosis | |
| CN113807431A (en) | An intelligent spindle state assessment method and system based on multi-source information fusion | |
| CN114048958A (en) | Tool wear prediction method and system based on deep convolutional residual shrinkage network | |
| CN109975702A (en) | A kind of DC gear decelerating motor product examine method based on recirculating network disaggregated model | |
| CN113326774A (en) | Machine tool energy consumption state identification method and system based on AlexNet network | |
| CN112763056A (en) | Method and system for online real-time monitoring and evaluation of state of numerical control machine tool system | |
| CN112364814A (en) | Photovoltaic robot speed detection method and system based on artificial intelligence image processing |
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 |