The working environment in which the equipment operates is basically a complex environment. The data obtained by the monitoring equipment is data mixed with noise interference. Obviously, the information we get is not all necessary. A large part of them are redundant, and the diagnosis of faults is useless. Therefore, it is necessary to remove these redundant information and simplify the fault information. Because the redundant information in the data is removed, the amount of data that the neural network needs to process is greatly reduced, the required network structure is less complicated, and the training time is also reduced.
In this paper, the rough set theory is applied to the front end of the neural network. The rough set is used to reduce the compressor fault samples, simplify the data, reduce the data processing volume of the neural network, realize the simplification of the neural network structure and the neural network fault in the compressor. Practical use in diagnosis.
Rough set-neural network intelligent hybrid compressor diagnostic system construction data preprocessing rough set theory only analyzes the discrete symbolic attribute values, and the symptom data detected in fault diagnosis is generally continuous quantity, therefore, in the application of coarse Before the set theory reduces the fault diagnosis decision system, it needs to discretely process the continuous attribute values. The so-called discretization of continuous attributes refers to dividing the attribute value of a numerical attribute into several sub-intervals, and replacing the original real value with this interval.
In this paper, the self-organizing map (SOM) neural network is used to discretize the continuous attribute values. Through the clustering function of the self-organizing map (SOM) neural network, the fault data is distributed in various types according to the specified number of clusters, and then used. The class name replaces the original data to achieve discretization, and the discrete result can objectively reflect the data distribution.
The implementation of this function is mainly applied to the creation function newsom of the SOM network of MATLAB and the calling format of the training function train<2>.Newsom function is as follows: Net=newsom(minmax(P),<22>); where P is the input The vector is the detected device data, minmax(P) specifies the maximum and minimum values ​​of the input vector, and <22> indicates that the competition layer of the created network is 22, and the network structure can be adjusted.
The data is then clustered and discrete using the designed network.
The reduction of data reduction data is realized by the distinct matrix method in the rough set theory. The distinct matrix method was proposed by A. Skowron in 1991, which can realize the reduction of fault data. The definition is as follows: If S=(U,A) is a knowledge expression system, A={a1,,am}, we use M(S) to represent the nn-order matrix (Cij), which is called the distinct matrix of S, where Cij ={aA:a(ui)a(uj)%ui,ujU%i,j=1,,n}Cij is composed of all the attributes that can distinguish the individual ui and uj. Obviously M(S) is symmetrical, and when i=1,, n, Ci=. So we only need the lower triangle to represent M(S), which corresponds to 1j. The time and space complexity of this algorithm varies exponentially with respect to the size of S, but it is the most effective way to calculate decision table reductions in many practices, such as workflow. Data Reduction Module Workflow 3 The extended neural network of the neural network can only correctly identify the type of fault that has been trained. Therefore, when a new fault type occurs in the equipment, it is necessary to retrain the network before it can respond to the new fault. Identification, which requires re-setting the network structure and erasing all the memory of the network, re-learning, obviously this is a defect of the single-sub-neural network. This system uses a multi-sub-neural network to make up for this deficiency. That is to say, when a new fault occurs, the original network structure is not changed, but a single fault diagnosis sub-network is reset and trained with a new fault sample. After completion, it is added to the original neural network. All the parameters of the original neural network remain unchanged, but a sub-network is added, thus avoiding repeated training of the neural network and saving time. The flow chart of the new sub-network of multi-child neural network is as follows. Multi-sub-neural network new sub-network increase process 4 rough set - neural network intelligent hybrid compressor diagnostic system implementation steps based on the rough set theory of the neural network compressor fault diagnosis system implementation block diagram as shown, the specific implementation steps are as follows: 1) Measure the process parameters of the operating conditions of the compressor to form a feature sample set; (2) quantize the continuous attribute values ​​with the SOM neural network; (3) establish a decision table with the discretized condition attributes and decision attributes and make Compatible; (4) Simplify conditional attributes with rough set theory: (a) compute the distinct matrix M(S) of the decision table system S; (b) compute the distinct function fM(S) associated with the distinct matrix; ) Calculate the minimum disjunction paradigm of the distinct function fM(S), which will give all reductions. (5) Construct a neural network on the reduced data set, using the optimized attribute combination, the quantized sample set corresponding to the simplified condition attribute as the input of the neural network, and the compressor fault type as the output of the network; Use various neural network learning algorithms to learn the network; (7) Repeat steps 5 and 6 until the effect of using this network classification can no longer be significantly improved; (8) Save the network and apply it to fault diagnosis. The steps of fault diagnosis using the trained neural network: (a) measuring the operating parameters of the compressor operating conditions; (b) screening the data for diagnosis according to the optimal decision table; (c) inputting the filtered data Train a good neural network for diagnosis; (d) output diagnostic results. The comparison of the rough set-neural network hybrid fault process diagnosis results is based on the fault data of four fault types of a certain type of compressor. The pure neural network and the rough neural network intelligent hybrid fault diagnosis system of this paper are used respectively. The fault data is learned and trained, and then the network structure of the two and the number of steps required for the training to reach convergence and the diagnostic error are compared.
Mould Standard Parts used in sheet stamping moulds, plastic moulds or FA automation equipment.Mold Standarts Parts include: punching needle, punch, guide pin, guide bushing, ejector pin, ejector sleeve, cylinder, steel ball sleeve, no oil guide sleeve, no oil feed slide, guide components.
Plastic Mould Standard Parts:
Single section ejector pin, double ejector pin,flat pin, locating pin, guide sleeve, ejector guide pin, guide bushing,return pin, angle pin, spring,stop pin, square location block, stereotaxic block group guide an auxiliary device, A, B, C type sprue bushing, locating ring type A, B, standard mechanical switch apparatus, small square auxiliary device,special sprue bushing for pin point gate,standard special sprue bushing for side gate, date indicating device, resin switch device, water stop switch and so on.
Sheet Stamping Standard Parts:
H-type straight punch, Type A two stage punch, Cutting edge forming punch, K type bud punch,B type guide punch, Type A baby punch, Punch, Die, Bushing, High speed steel round rod,
Ultra-fine particle tungsten steel round rod, Powder high speed steel round rod, Lifting pin,
Floating pin, Stop payment screw, Locating pin (fixed pin), Contour sleeve, Guide post,
Guide bush, Precision chrome-plated guide column, Precision grade copper-titanium alloy guide sleeve, Self-lubricating guide sleeve, Internal guide column assembly, Sliding guide column assembly for die base, Ball guide column assembly for die holder, Detachable ball guide column assembly, External guide column assembly, Ball sleeve (cage), Independent guide column, Hexagon screw, Contour screws, etc.
Sheet Metal Stand,Stand Metal Parts,Mould Stand Spare Parts,Injection Mould Spare Parts
Ningbo Expert Mould Plastic Co., Ltd. , https://www.prototypesmaker.com