CN115114971A - Logistics sorting equipment fault diagnosis method based on BP neural network - Google Patents
Logistics sorting equipment fault diagnosis method based on BP neural network Download PDFInfo
- Publication number
- CN115114971A CN115114971A CN202111444375.3A CN202111444375A CN115114971A CN 115114971 A CN115114971 A CN 115114971A CN 202111444375 A CN202111444375 A CN 202111444375A CN 115114971 A CN115114971 A CN 115114971A
- Authority
- CN
- China
- Prior art keywords
- fault
- layer
- fault diagnosis
- output
- input
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 58
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 210000002569 neuron Anatomy 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 7
- 238000012423 maintenance Methods 0.000 claims description 5
- 241000252254 Catostomidae Species 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000003672 processing method Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 2
- 238000013399 early diagnosis Methods 0.000 abstract description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000004422 calculation algorithm Methods 0.000 description 10
- 235000019504 cigarettes Nutrition 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 239000008358 core component Substances 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 241000208125 Nicotiana Species 0.000 description 1
- 235000002637 Nicotiana tabacum Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000000306 component Substances 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a logistics sorting equipment fault diagnosis method based on a BP neural network, which comprises the following steps of counting dominant faults of equipment, and establishing a fault library; counting historical data of the logistics sorting equipment with the dominant faults to form a fault library; establishing a fault diagnosis model based on a BP neural network structure; acquiring a plurality of groups of data of the logistics sorting equipment under the fault condition as the input of a fault diagnosis model according to the input parameter type, and performing a plurality of times of iterative training on all weights and thresholds of the fault diagnosis model by using MATLAB programming; verifying the accuracy; and (4) fault diagnosis, namely periodically counting the actual operation values of various parameters, inputting the actual operation values into a system carrying a fault diagnosis model, and automatically outputting the fault diagnosis result of the equipment by the system. The beneficial effects of the invention are as follows: the device is convenient to use and high in detection accuracy, can find the hidden fault of the device, achieves early diagnosis, early finding, early treatment and early solution, and avoids more serious faults.
Description
Technical Field
The invention relates to the field of fault diagnosis, in particular to a logistics sorting equipment fault diagnosis method based on a BP neural network.
Background
The logistics cigarette sorting equipment is the most main tool for realizing cigarette sorting and distribution, the normal operation of the sorting equipment is a material basis for ensuring that cigarettes are distributed to customers at regular time and on a point basis, along with the advance of high-precision technology, the continuous development of computer measurement technology and signal processing technology, the requirement of an information control system is more and more complex, so that the fault diagnosis of a mechanical arm is more and more difficult, the traditional fault tree analysis method which is widely applied cannot completely meet the requirement of equipment detection and development, although the fault tree analysis method can find out the links of equipment faults and the dominant faults of the equipment through gradual refinement, decomposition and development, some current technologies and methods also show limitation. Firstly, in the process of fault tree analysis, expert comprehensive judgment is needed, and diagnosis rules are easily limited by the field of individual expert knowledge and deviate from core problems; secondly, different experts have different opinions and descriptions on the same problem, so that data is difficult to acquire; thirdly, the fault tree analysis can only detect the dominant fault, and the hidden fault is difficult to be checked;
with the development of logistics equipment fault diagnosis intellectualization, some researchers put forward to combine data mining technology with fault diagnosis and apply a matrix weighting association rule algorithm to carry out intelligent diagnosis on equipment faults, and the basic assumption of the association rule is that fault factors of all parts in a database have the same importance and are uniformly distributed, but in actual fault diagnosis, different fault factors have different fault contribution degrees and different occurrence frequencies, so the association rule algorithm cannot be completely applied.
The artificial neural network algorithm can solve the complex problem of uncertain factors, can identify core fault factors, gives different weights, analyzes and compares newly-added data, automatically brings the newly-added data into a fault library, can judge fuzzy causal relationships existing in input and output, and finally accurately judges the fault state, and has strong universality and obvious advantages. In the actual work, the abnormal shape cigarette sorting equipment of enterprise operation is unstable, the down time is long, the fault rate is high, and the operating personnel examines to the point of commodity circulation facility equipment, maintenance and maintenance stop clean, twist on the screw on the surface, the key spare part and the weighing parameter of core component do not discerned, can not judge the fault condition of equipment, wait that equipment spare part is bad, could discover and change, the degree of wear of equipment has not only accelerated, equipment trouble shut down risk has still been increased, both influenced the normal letter sorting delivery of abnormal shape cigarette, still cause personnel overload work scheduling problem. Therefore, it is very important to establish an intelligent equipment fault diagnosis and analysis model.
Disclosure of Invention
The invention aims to solve the problems that the fault of the logistics special-shaped cigarette sorting equipment is difficult to find, maintain and maintain, the fault state of the equipment cannot be identified and diagnosed manually, and the equipment cannot be scientifically and accurately maintained.
The invention is realized by the following technical scheme:
the invention discloses a logistics sorting equipment fault diagnosis method based on a BP neural network, wherein the method comprises the following steps: the method comprises the following steps:
counting the dominant faults of the equipment, and establishing a fault library; counting the historical data of the explicit faults of the logistics sorting equipment to form a fault library, wherein the content of the fault library comprises fault content, fault types, downtime, reasons for generation, solutions, maintenance cost and measurement parameters;
analyzing the fault library and determining the input parameter type of the BP neural network structure input layer;
establishing a fault diagnosis model based on a BP neural network structure, wherein the fault diagnosis model adopts a three-layer network structure and is an input layer, a hidden layer and an output layer respectively; the number of neurons of the input layer is equal to the number of input parameter types of the input layer of the BP neural network structure, the number of neurons of the output layer is limited to three, and then the number of neurons of the hidden layer is calculated;
according to the input parameter type, acquiring a plurality of groups of data of the logistics sorting equipment under the fault condition as the input of a fault diagnosis model, taking the training frequency as 1000, the training target as 0.01 and the minimum value of network errors as a fitness function, and performing iterative training on all weights and threshold values of the fault diagnosis model for a plurality of times by using MATLAB programming;
verifying the accuracy, randomly counting data parameters of the operation of the five groups of mechanical arms as input values of the fault diagnosis model to verify sample data, and if the result is verified to be within an error range, executing the next step, and if the result is outside the error range, repeating the previous step;
and (3) fault diagnosis, namely periodically counting actual operation values of various parameters, inputting the actual operation values into a system carrying a fault diagnosis model, and automatically outputting equipment fault diagnosis results by the system, wherein the output results comprise: (1, 0, 0) no fault, (0,1,0) recessive fault, (0, 0, 1) dominant fault.
Preferably, in the fault diagnosis model established based on the BP neural network structure, the hidden layer uses an S-type tangent function, and the output layer uses an S-type logarithmic function.
Preferably, the iterative computation process of the fault diagnosis model includes: forward propagation: information enters a network from an input layer, is subjected to information transformation through a hidden layer, and generates an output signal at an output end; in the process, the weight value of the network is kept unchanged, and the state of each layer of neurons is only related to the state of the neurons in the previous layer; if the actual output value and the expected value are larger than the expected error, switching to the error signal reverse propagation; assuming that X is an input vector and the number of input nodes is n; y is the corresponding output vector, the number of output nodes is m, then:
X＝(x1,x2,…,xk,…,xn),1≤k≤n
T＝(T1,T2,…,Tm,…,Tq),1≤m≤q
the steps of forward propagation are as follows:
the output value of the hidden layer node is calculated. Taking hidden layer node j as an example, let the number of hidden layer nodes be p, w 1kj For connection weight, the threshold value of the node j is theta 1j Input value is i 1j Output value of h j Then, then
i 1j ＝∑W 1kj x k ，1≤j≤p
h j ＝f(i 1j +θ 1j )
Wherein the activation function f is a sigmoid function;
calculating an output value of an output layer node; taking the output layer node m as an example, let the input value of the node m be i 2m Output value of o m The threshold value is theta 2m Then, then
i 2m ＝∑w 2jm h j ，1≤m≤q
o m ＝f(i 2m +θ 2m )
And (3) reverse propagation: inputting error signals at an output end, transmitting the error signals forwards through the hidden layer in a backward transmission mode, averagely distributing the error signals to all units of each layer, and adjusting the connection weight of each unit layer by using the signal error of each layer of unit, thereby reducing the error signals; the steps of the reverse propagation are as follows:
firstly, calculating an output error of an output layer node and detecting; learning value o of output layer node m m Target output value T of learning sample m The error of (c) is:
ε m ＝|o m -T m |
let ε be m Maximum learning error allowed, if max (ε) m )≤ε 0 If not, adjusting network parameters and carrying out original learning again until all learning samples meet the conditions;
recalculating the learning error, the learning error d of nodes m and j 2m And d 1j Can be obtained by the following formula:
d 2m ＝o m (1-o m )(o m -T m )
d 1j ＝h j (1-h j )∑w 2jm d 2m
and finally, adjusting network parameters. Setting the weight value of the time t +1 as a new weight value after the adjustment, the following steps are performed:
w 2jm (t+1)＝w 2jm (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1)]
w 1kj (t+1)＝w 1kj (t)+ηd 1j x k +α[w 1kj (t)-w 1kj (t-1)]
in the formula, eta belongs to [0, 1] as the learning rate, and alpha belongs to [0, 1] as the momentum factor;
correction thresholds θ 1 and θ 2:
θ 2m (t+1)＝θ 2m (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1) θ 1j (t+1)＝θ 1j (t)+ηd1 j+α[θ 1j (t)-θ 1j (t-1)]。
preferably, the method for calculating the neuron number of the hidden layer comprises the following steps: using the formula: n is 2 ＝2×n 1 +1 calculation of n in the formula 1 As the number of input layer neurons, n 2 The number of layer neurons is implied.
Preferably, each group of the plurality of groups of data of the logistics sorting equipment under the fault condition respectively comprises parameters of vacuum pump negative pressure, vacuum pump positive pressure, whether suckers are parallel or not, anti-collision air cylinder air pressure, action times (times/minutes), speed, mechanical arm coordinate offset degree, whether a PLC electric wire is broken or not, times that a laser limit detection error is less than 15mm, and whether a connecting rod processing method is appropriate or not; the parameter of whether the suckers are parallel or not is represented by the number 1, and the number 0 represents non-parallelism; the parameter of whether the PLC wire is broken is represented by a numeral 1 to indicate that the PLC wire is not broken and a numeral 0 to indicate that the PLC wire is broken; the parameters of whether the connecting rod processing method is proper are represented by 1, and improper are represented by 0.
Preferably, the fault diagnosis model based on the BP neural network structure is n 1 -n 2 -n 3 I.e. the input layer n 1 Number of neurons, hidden layer n 2 Number of individual neurons, output layer n 3 Number of individual neurons, n 3 Where the number of neurons is the number of nodes in the layer, n is the total number 1 ×n 2 +n 2 ×n 3 A weight value, n 2 +n 3 A threshold value.
The invention has the beneficial effects that: the device is convenient to use and high in detection accuracy, can find the hidden fault of the device, achieves early diagnosis, early discovery, early treatment and early solution, and avoids more serious faults.
Detailed Description
The invention will be further illustrated with reference to specific embodiments:
example (b): a logistics sorting equipment fault diagnosis method based on a BP neural network is disclosed, wherein: 1, counting dominant faults of equipment, and establishing a fault library; counting the historical data of the explicit faults of the logistics sorting equipment to form a fault library, wherein the content of the fault library comprises fault content, fault types, downtime, reasons for generation, solutions, maintenance cost and measurement parameters; 2. analyzing the fault library and determining the input parameter type of the BP neural network structure input layer; 3. establishing a fault diagnosis model based on a BP neural network structure, wherein the fault diagnosis model adopts a three-layer network structure and is an input layer, a hidden layer and an output layer respectively; the number of neurons of the input layer is equal to the number of input parameter types of the input layer of the BP neural network structure, the number of neurons of the output layer is limited to three, and then the number of neurons of the hidden layer is calculated; 4. according to the input parameter type, collecting a plurality of groups of data of the logistics sorting equipment under the fault condition as the input of a fault diagnosis model, taking the training frequency as 1000, the training target as 0.01 and the minimum value of network errors as a fitness function, and performing iterative training on all weights and threshold values of the fault diagnosis model for a plurality of times by using MATLAB programming; 5. carrying out accuracy verification, randomly counting data parameters of the operation of the five groups of mechanical arms as input value verification sample data of the fault diagnosis model, verifying the result, executing the next step if the result is within an error range, and repeating the previous step if the result is outside the error range; 6. and (3) fault diagnosis, namely periodically counting the actual operation values of various parameters, inputting the actual operation values into a system carrying a fault diagnosis model, and automatically outputting equipment fault diagnosis results by the system, wherein the output results comprise: (1, 0, 0) no fault, (0,1,0) recessive fault, (0, 0, 1) dominant fault.
When in work: taking the fault diagnosis of the special-shaped cigarette sorting equipment in the logistics of the tobacco company as an example, in order to realize the early diagnosis, the early discovery, the early treatment and the early solution, the method comprises the following specific working steps:
counting dominant faults of the equipment to form a fault library; the faults generated by the mechanical arm are counted to form a fault library, and due to the large data volume, only part of data is selected in the embodiment, as shown in the following table:
fault library for 7-9 month statistics in 2018
Through the establishment of a fault library, the core components with faults of the mechanical arm are further analyzed and confirmed to be concentrated on components such as a vacuum pump, a sucker, a mechanical arm coordinate and an anti-collision cylinder, and related key parameters comprise positive pressure, negative pressure, coordinate errors and the like.
Establishing a fault diagnosis model; the BP neural network learning algorithm takes actual parameters of equipment operation as input, trains samples, finds out data intrinsic rules through multiple times of training of actual operation parameter values, establishes a diagnosis model, and finally outputs three fault states (namely (1, 0, 0) representing no fault) by a computer; (0,1,0) represents a latent fault; (0, 0, 1) represents a dominant fault.
In the embodiment, 29 groups of data of 10 key parameters under the fault mode of core parts such as a vacuum pump, a sucking disc and an anti-collision cylinder of the mechanical arm are collected to serve as training samples, and the BP network is trained. The following table is statistical training sample data.
BP network input parameter statistical table
The model algorithm for fault diagnosis adopts a three-layer network structure, a hidden layer uses an S-type tangent function, and an output layer uses an S-type logarithmic function. Using the formula: n is 2 ＝2×n 1 +1 to calculate the number of hidden layer neurons in the neural network, where n 1 Is an input layerNumber of neurons, n 2 The number of layer neurons is implied. Since there are 10 input parameters and 3 output parameters in a sample, the BP neural network structure is set to 10-21-3, that is, 10 nodes in an input layer, 21 nodes in an implicit layer, and 3 nodes in an output layer, there are 10 × 21+21 × 3 ═ 273 weights, 21+3 ═ 24 thresholds, the total number of all parameters to be optimized is 297, the training frequency of the network is 1000, the training target is 0.01, and the minimum value of the error of the network is selected as a fitness function. Through MATLAB programming training iteration of a BP neural network, from the 1 st iteration and the 19 th iteration, the training errors are respectively 0.309 and 0.0000000856 and almost approach to 0, and the fact that actual data basically accord with model data is shown.
The mathematical calculation process of the iterative training is as follows: forward propagation: information enters a network from an input layer, is subjected to information transformation through a hidden layer, and generates an output signal at an output end; in the process, the weight value of the network is kept unchanged, and the state of each layer of neuron is only related to the state of the neuron at the previous layer; if the actual output value and the expected value are larger than the expected error, switching to the error signal for reverse propagation; assuming that X is an input vector and the number of input nodes is n; y is the corresponding output vector, the number of output nodes is m, then:
X＝(x1，x2，…，xk，…，xn)，1≤k≤n
T＝(T1，T2，…，Tm，…，Tq)，1≤m≤q
the steps of forward propagation are as follows:
and calculating the output value of the hidden layer node. Taking hidden layer node j as an example, let the number of hidden layer nodes be p, w 1kj For connection weight, the threshold value of the node j is theta 1j Input value is i 1j Output value of h j Then, then
i 1j ＝∑w 1kj x k ，1≤j≤p
h j ＝f(i 1j +θ ij )
Wherein the activation function f is a sigmoid function;
calculating an output value of an output layer node; taking node m of the output layer as an example, let the input value of node m be i 2m Output value of o m The threshold value is theta 2m Then, then
i 2m ＝∑w 2jm h j ，1≤m≤q
o m ＝f(i 2m +θ 2m )
And (3) reverse propagation: inputting error signals at an output end, transmitting the error signals forwards through the hidden layer in a backward transmission mode, averagely distributing the error signals to all units of each layer, and adjusting the connection weight of each unit layer by using the signal error of each layer of unit, thereby reducing the error signals; the steps of the reverse propagation are as follows:
firstly, calculating an output error of an output layer node and detecting; learning value o of output layer node m m Target output value T of learning sample m The error of (2) is:
ε m ＝|o m -T m |
let ε be m Maximum learning error allowed, if max (ε) m )≤ε 0 If not, adjusting network parameters and carrying out original learning again until all learning samples meet the conditions;
recalculating the learning error, the learning error d of nodes m and j 2m And d 1j Can be obtained by the following formula:
d 2m ＝o m (1-o m )(o m -T m )
d 1j ＝h j (1-h j )∑w 2jm d 2m
and finally, adjusting network parameters. Setting the weight of the time t +1 as a new weight after the adjustment, the following steps are performed:
w 2jm (t+1)＝w 2jm (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1)]
w 1kj (t+1)＝w 1kj (t)+ηd 1j x k +α[w 1kj (t)-w 1kj (t-1)]
in the formula, eta belongs to [0, 1] as the learning rate, and alpha belongs to [0, 1] as the momentum factor;
correction thresholds θ 1 and θ 2:
θ 2m (t+1)＝θ 2m (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1) θ 1j (t+1)＝θ 1j (t)+ηd1j+α[θ 1j (t)-θ 1j (t-1)]。
regarding the BP neural network, in 1986, Rumelhant and McClelland propose an Error Back Propagation (Error Back Propagation) learning algorithm of a multilayer feedforward network, which is called BP algorithm for short, and the algorithm is a Back-push learning algorithm of the multilayer network. The three-layer network is a typical BP neural network structure, and comprises an input layer, a hidden layer (intermediate layer) and an output layer, wherein the node number of each layer is set according to the requirements of a specific problem. The training process of the BP neural network is mainly completed through two stages of forward propagation of signals and backward propagation of errors. This is a gradient descent algorithm, which can make the error of each connection weight value of the neural network continuously decrease.
Verifying experimental analysis; in order to verify the trained network, the enterprise randomly counts the data of the five groups of mechanical arm operation as the verification sample data of the network, and the verification result is shown as the table:
detecting sample detection data
Test sample validation results
Through verification, the actual output of the five groups of data is basically consistent with the expected output within the error allowable range, and the result shows that the BP neural network is successfully constructed, the accuracy of the BP neural network in mechanical arm fault diagnosis is high, and the operation is stable.
The system with the BP neural network operation fault diagnosis model enables personnel in the logistics line to be free from knowing the principle of a statistical method, only needs to regularly count actual operation values of various parameters and input the actual operation values into the system, and can automatically output equipment fault diagnosis results, such as (1, 0, 0) no fault, (0,1,0) hidden fault, and (0, 0, 1) dominant fault; the applicant monitors the operation data of the mechanical arm in late 9 th of 2018 by using the method of the embodiment, finds that the air pressure value of the mechanical arm is abnormal in 26 th of 9 th of 2018, the fault detection is in a hidden fault state, the team member checks in time, finds that the hose of the mechanical arm leaks air, and rapidly maintains the corresponding parts, so that more serious faults are avoided, the equipment detection accuracy rate reaches 100%, and the countermeasure implementation is effective.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments or portions thereof without departing from the spirit and scope of the invention.
Claims (6)
1. A logistics sorting equipment fault diagnosis method based on a BP neural network is characterized in that: the method comprises the following steps:
counting the dominant faults of the equipment, and establishing a fault library; counting the historical data of the explicit faults of the logistics sorting equipment to form a fault library, wherein the content of the fault library comprises fault content, fault types, downtime, reasons for generation, solutions, maintenance cost and measurement parameters;
analyzing the fault library, and determining the input parameter type of the BP neural network structure input layer;
establishing a fault diagnosis model based on a BP neural network structure, wherein the fault diagnosis model adopts a three-layer network structure and is an input layer, a hidden layer and an output layer respectively; the number of neurons of the input layer is equal to the number of input parameter types of the input layer of the BP neural network structure, the number of neurons of the output layer is limited to three, and then the number of neurons of the hidden layer is calculated;
according to the input parameter type, collecting a plurality of groups of data of the logistics sorting equipment under the fault condition as the input of a fault diagnosis model, taking the training frequency as 1000, the training target as 0.01 and the minimum value of network errors as a fitness function, and performing iterative training on all weights and threshold values of the fault diagnosis model for a plurality of times by using MATLAB programming;
carrying out accuracy verification, randomly counting data parameters of the operation of the five groups of mechanical arms as input value verification sample data of the fault diagnosis model, verifying the result, executing the next step if the result is within an error range, and repeating the previous step if the result is outside the error range;
and (3) fault diagnosis, namely periodically counting actual operation values of various parameters, inputting the actual operation values into a system carrying a fault diagnosis model, and automatically outputting equipment fault diagnosis results by the system, wherein the output results comprise: (1, 0, 0) no fault, (0,1,0) recessive fault, (0, 0, 1) dominant fault.
2. The method for diagnosing the faults of the logistics sorting equipment based on the BP neural network as claimed in claim 1, wherein the method comprises the following steps: according to the fault diagnosis model established based on the BP neural network structure, the hidden layer uses an S-type tangent function, and the output layer uses an S-type logarithmic function.
3. The method for diagnosing the fault of the logistics sorting equipment based on the BP neural network as claimed in claim 1 or 2, wherein: the iterative calculation process of the fault diagnosis model comprises the following steps:
forward propagation: information enters a network from an input layer, information transformation is carried out through a hidden layer, and an output signal is generated at an output end; in the process, the weight value of the network is kept unchanged, and the state of each layer of neurons is only related to the state of the neurons in the previous layer; if the actual output value and the expected value are larger than the expected error, switching to the error signal reverse propagation; assuming that X is an input vector and the number of input nodes is n; y is the corresponding output vector, the number of output nodes is m, then:
X＝(x1，x2，…，xk，…，xn)，1≤k≤n
T＝(T1，T2，…，Tm，…，Tq)，1≤m≤q
the forward propagation steps are as follows:
the output value of the hidden layer node is calculated. Taking hidden layer node j as an example, let the number of hidden layer nodes be p, w 1kj For connection weight, the threshold value of the node j is theta 1j Input value is i 1j Output value of h j Then, then
i 1j ＝∑W 1kj x k ，1≤j≤p
h j ＝f(i 1j +θ 1j )
Wherein the activation function f is a sigmoid function;
calculating an output value of an output layer node; taking node m of the output layer as an example, let the input value of node m be i 2m Output value of o m The threshold value is theta 2m Then, then
i 2m ＝∑w 2jm h j ，1≤m≤q
o m ＝f(i 2m +θ 2m )
And (3) reverse propagation: inputting error signals at an output end, then transmitting the error signals forwards through the hidden layer in a back transmission mode, averagely distributing the error signals to all units of each layer, and adjusting the connection weight of each unit layer by using the signal error of each layer of the obtained units so as to reduce the error signals; the steps of the reverse propagation are as follows:
firstly, calculating an output error of an output layer node and detecting; learning value o of output layer node m m Target output value T of learning sample m The error of (c) is:
ε m ＝|o m -T m |
let ε be m Maximum learning error allowed, if max (ε) m )≤ε 0 If not, adjusting network parameters and carrying out original learning again until all learning samples meet the conditions;
recalculating the learning error, the learning error d of nodes m and j 2m And d 1j Can be obtained by the following formula:
d 2m ＝o m (1-o m )(o m -T m )
d 1j ＝h j (1-h j )∑w 2jm d 2m
and finally, adjusting network parameters. Setting the weight value of the time t +1 as a new weight value after the adjustment, the following steps are performed:
W 2jm (t+1)＝w 2jm (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1)]
w 1kj (t+1)＝w 1kj (t)+ηd 1j X k +α[w 1kj (t)-w 1kj (t-1)]
in the formula, eta belongs to [0, 1] as learning rate, and alpha belongs to [0, 1] as momentum factor;
correction thresholds θ 1 and θ 2:
θ 2m (t+1)＝θ 2m (t)+ηd 2m h j +α[w 2jm (t)-w 2jm (t-1)
θ 1j (t+1)＝θ 1j (t)+ηd1j+α[θ 1j (t)-θ 1j (t-1)]。
4. the method for diagnosing the faults of the logistics sorting equipment based on the BP neural network as claimed in claim 1, wherein the method comprises the following steps: the method for calculating the neuron number of the hidden layer comprises the following steps:
using the formula: n is 2 ＝2×n 1 +1 calculation of n in the formula 1 As the number of input layer neurons, n 2 Is implied for the number of layer neurons.
5. The method for diagnosing the fault of the logistics sorting equipment based on the BP neural network as claimed in claim 1, wherein: the logistics sorting equipment comprises a plurality of groups of data under the fault condition, wherein each group of data respectively comprises parameters of vacuum pump negative pressure, vacuum pump positive pressure, whether suckers are parallel or not, anti-collision cylinder air pressure, action times (times/minutes), speed, mechanical arm coordinate offset degree, PLC (programmable logic controller) wire breakage, times of laser limit detection error smaller than 15mm and whether a connecting rod processing method is proper or not;
the parameter of whether the suckers are parallel is represented by the number 1, and the number 0 represents non-parallelism; the parameter of whether the PLC wire is broken is represented by a numeral 1 to indicate that the PLC wire is not broken, and is represented by a numeral 0 to indicate that the PLC wire is broken; the parameters of whether the connecting rod processing method is proper are represented by 1, and improper are represented by 0.
6. The method for diagnosing the faults of the logistics sorting equipment based on the BP neural network as claimed in claim 1, wherein the method comprises the following steps: the fault diagnosis model based on the BP neural network structure is n 1 -n 2 -n 3 I.e. the input layer n 1 Number of neurons, hidden layer n 2 Number of nerve cells, output layer having n 3 Number of individual neurons, n 3 Where the number of neurons is the number of nodes in the layer, n is the total number 1 ×n 2 +n 2 ×n 3 A weight value, n 2 +n 3 A threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111444375.3A CN115114971A (en) | 2021-11-30 | 2021-11-30 | Logistics sorting equipment fault diagnosis method based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111444375.3A CN115114971A (en) | 2021-11-30 | 2021-11-30 | Logistics sorting equipment fault diagnosis method based on BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115114971A true CN115114971A (en) | 2022-09-27 |
Family
ID=83325025
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111444375.3A Pending CN115114971A (en) | 2021-11-30 | 2021-11-30 | Logistics sorting equipment fault diagnosis method based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115114971A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115796237A (en) * | 2022-12-07 | 2023-03-14 | 北京石油化工学院 | ICSA-VS-bpNet-based armored vehicle chassis engine fault prediction method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090658A (en) * | 2017-12-06 | 2018-05-29 | 河北工业大学 | Arc fault diagnostic method based on time domain charactreristic parameter fusion |
| CN110766143A (en) * | 2019-10-31 | 2020-02-07 | 上海埃威航空电子有限公司 | Equipment fault intelligent diagnosis method based on artificial neural network |
-
2021
- 2021-11-30 CN CN202111444375.3A patent/CN115114971A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090658A (en) * | 2017-12-06 | 2018-05-29 | 河北工业大学 | Arc fault diagnostic method based on time domain charactreristic parameter fusion |
| CN110766143A (en) * | 2019-10-31 | 2020-02-07 | 上海埃威航空电子有限公司 | Equipment fault intelligent diagnosis method based on artificial neural network |
Non-Patent Citations (1)
| Title |
|---|
| 冯玉芳: "基于BP神经网络的故障诊断模型研究", 《计算机工程与应用》, vol. 55, no. 6, 31 March 2019 (2019-03-31), pages 1 - 7 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115796237A (en) * | 2022-12-07 | 2023-03-14 | 北京石油化工学院 | ICSA-VS-bpNet-based armored vehicle chassis engine fault prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020077672A1 (en) | Method and device for training service quality evaluation model | |
| CN108197648A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models | |
| CN112381180B (en) | Power equipment fault monitoring method based on mutual reconstruction single-class self-encoder | |
| CN113704075B (en) | Fault log-based high-performance computing system fault prediction method | |
| Chang et al. | A two-stage neural network approach for process variance change detection and classification | |
| CN113255848A (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN111199252A (en) | Fault diagnosis method for intelligent operation and maintenance system of power communication network | |
| CN111723850A (en) | Automatic verification equipment state evaluation method based on intelligent inspection system | |
| CN111879349A (en) | Sensor data deviation self-adaptive correction method | |
| CN111798095A (en) | Power cable state evaluation method based on neural network | |
| CN113923104A (en) | Network fault diagnosis method, equipment and storage medium based on wavelet neural network | |
| CN113033078B (en) | Construction method, system and early warning method of fault early warning model of relay protection equipment | |
| CN115114971A (en) | Logistics sorting equipment fault diagnosis method based on BP neural network | |
| CN113887729A (en) | Fault diagnosis method for low-voltage power line carrier communication system | |
| CN114527714A (en) | Workshop dynamic scheduling method based on digital twin and disturbance monitoring | |
| CN110443481B (en) | Power distribution automation terminal state evaluation system and method based on hybrid K-nearest neighbor algorithm | |
| CN114970309A (en) | Thermal power equipment state early warning evaluation method | |
| CN114020598A (en) | Method, device and equipment for detecting abnormity of time series data | |
| CN114189047A (en) | False data detection and correction method for active power distribution network state estimation | |
| Ghanaatiyan et al. | Multi-objective economic-statistical design of VSSI-MEWMA-DWL control chart with multiple assignable causes | |
| CN105741184B (en) | Transformer state evaluation method and device | |
| CN117234785A (en) | Centralized control platform error analysis system based on artificial intelligence self-query | |
| CN116776209A (en) | Method, system, equipment and medium for identifying operation state of gateway metering device | |
| CN113724211B (en) | Fault automatic identification method and system based on state induction | |
| CN112748663B (en) | Wind power torque fault-tolerant control method based on data-driven output feedback |
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 |