Detection of Coal Mine Spontaneous Combustion by Fuzzy Inference System

Sept 2006J.China Univof Mining& Tech ( English Edition)oL. 16 No. 3Detection of Coal Mine Spontaneous Combustion by Fuzzy Inference SystemSUN Ji-ping, SONG Shu, MA Feng-ying", ZHANG Ya-liI Institute of Information Engineering, China University of Mining Technology, Beijing 100083, chinaCollege of Electronic Information and Control Engineering, Shandong Institute of light Industry, Jinan, Shandong 250100, Chinapleing 100038, ChinaAbstract: The spontaneous combustion is a smoldering process and characterized by a slow burning speed and a longduration. Therefore, it is a hazard to coal mines. Early detection of coal mine spontaneous combustion is quite difficultecause of the complexity of different coal mines. And the traditional threshold discriminance is not suitable forspontaneous combustion detection due to the uncertainty of coalmine combustion Restrictions of the single detectionmethod will also affect the detection precision in the early time of spontaneous combustion. Although multiple detectionmethods can be adopted as a complementarity to improve the accuracy of detection, the synthesized method willlicacy of critericing it difficult to estimate the combustion. To solve this problem, a fuzzyinference system based on CRI(Compositional Rule of Inference) and fuzzy reasoning method FITA(First InferThen Aggregate)are presented. And the neural network is also developed to realize the fuzzy inference systemFinally, the effectiveness of the inference system is demonstrated by means of an experimentKey words: spontaneous combustion; fuzzy inference system; CRI; FITA; neural networkCLC number: TD 75: TP 183: X9361 IntroductionIn the early detection of coalmines spontaneouscombustion, the information acquired by all sorts ofSpontaneous combustion is one of the disasters detection means is uncertain. And so the detectionin coalmines. It may potentially induce other coal results cannot tally with thresholds preestablishedmine accidents and so it's necessary to detect them in CRI(Compositional Rule of Inference)sed inthe early stages. The spontaneous combustionthe fuzzy inference system to suit multi-detectionsmoldering process wIth a slow burning speed and a methodlong developing time. All these make the spontaneousFuzzy inference based on the artificial neuralcombustion difficult to detect in the early timenetworks was presented to realize the identification ofThe threshold method is usually adopted in the spontaneous combustion in coalminedetecting coalmine spontaneous combustion. TheThis paper is organized as follows. In Section IL,temperature, temperature variation, the concentration fuzzy reasoning and fuzzy inference system as wellof index gas and its variation are frequently the compositional rule of inference is introducedconsidered as the chief combustion detection signals. Section Ill presents the fuzzy system based onNo matter which detection signal of combustion artificial neural networks. Section IV presents anis adopted, the detection results are always different illustrative experiment. And the conclusion is givenin reliability. The early detection results of coalthe last sectionthreshold method. Actually the detection result of 2 Fuzzy Inference System and Fuzzyspontaneous combustion is not simple“Yes”"or“No”,Reasoningbut different in stage or degree in combustionOne of the important research topics ofTraditional detection technology is not sensitive rule-based fuzzy system is the management ofto those faint signals of spontaneous combustion in uncertainty. The system must have an ability tothe early time. Besides the spontaneous combustion perform fuzzy reasoning especially in an uncertainorientation of coal, the complex conditionsenvironment, just like spontaneous combustion in thedifferent coalmines also affect the combustion. Thus coalmine. Fuzzy set and the fuzzy logic are used inthe threshold of different signal cannot be a the systemquantitative index by classical mathematics meansFuzzy中国煤化工tant part ofReceived 10 April 2006: accepted 15 May 2006Project 20050290010 supported by the Doctoral Foundation of Chinese Education Ministry andTHCNMHGor HichTechnique Research DevelopmentCorrespondingauthorTel:+86-10-62331011:E-mailaddressshu_song@163.comSun Ji-ping et al.Detection of Coal Mine Spontaneous Combustion by Fuzzy Inference Systemfuzzy system. It's one of the uncertainty reasoningAnother method fati is to assemble the rules atmethods. It works as: for any observation facts, a first, then infers the consequence by CRI. The FITApossibly imprecise result can be inferred by the was adopted in the paper because each detectionsystem through a set of rules. The main function of method of spontaneous combustion can obtain afuzzy reasoning is that a result can be obtained even reasonable consequence of combustionthe observation facts dont match the antecedentportion of the rule precisely3 Fuzzy Weighted Inference SystemFor a given ruleIF X is AKArtificial neural networks can obtain knowledgeThen Y is Bx K=1, 2, ,n(1) expressed by data through learning and training. Butwhere X is the input variable, Ak is a linguistic value the knowledge existing in network weights is difficultsuch as"small"and"large" y is the output variable,to understand. The character of a fuzzy system is thatBk a consequent linguistic value described bit can express logic directly. It has the ability to dealmembership that iswith uncertain information but it is not an activeAk(x) and lBK(y)∈O,1learning system like neural network. The knowledgethat the artificial neural networks have comes fromthe existing database and the rules used for reasoningFor an observation factcome from the expert database. The more the rules inX Is A(2) the system, the more complex the system will beBy fuzzy reasoning, a consequence can be which restrict its'application. The combination ofobtainedfuzzy systems with neural networks leads to theYis B(3) generation of fuzzy neural networks 9-10)(ENN).ItCRI(Compositional Rule of Inference)has the self-learning ability andwidely used fuzzy reasoning method -, which was knowledge clearly. A fuzzy system based on neuralput forward by L A ZAdeh in 1973. For a given rule networks and Cri reasoning method is developed inIF X is A then y is B and an observation X is at', combustionthis paper to detect the coalmine spontaneousa consequence can be obtained under the CriWhen one rule is adopted, an available 3. 1 Construction fuzzy neural systemsthe rule matches the observation precisely. BecauseFor an n-input-one-output fuzzy system ,itfuzzy reasoning with one rule can only change thhas the common form as followsmembership of the rules rear part, a right reasoningIf x, is A'. x, is A, x, is A/, then y isconsequence cannot be obtained. Thus, rules should j=1, 2,, K. The output of the fuzzy system wasbe combined before CRi is appliedthe consequence inferred by CrI e single rule (1)calculated as folloFor the observation(2)and∑Hy2(y)=(T(A(x),A→B(x,y)(4)∑where HA(x ),Bg(,),UA(x, )and AB, (y )E[0,II1,2,1=p(x1)(x2)When several rules like (1)are given, theThe Cri was applied in the system of detectingfollowing method FITA+(First Infer Then spontaneous combustion. Rules of each detectionAggregate) can be used to solve the problemmethod(such as thermometry, audiometry, IR, etc.First, infer consequence of each rule and factwere set down first. Neural networkssed toaccomplish the whole process of the compositionalWg()=V(T(HA (x ) WAx -B,(x;, y )))(5) inference. There are K+1 neural networksthesystem. The value of the k depends on the number ofdetection methods adopted and it determines theThen aggregate the results abovecomplexity of the fuzzy, system. The model of(y)=B1(y)④pg2(y)④…,④pgn(y)(6network is shown in FigNNI,., NNK are used to express the reasoningwhere e∈(V,∧) is an arithmetic operatorsresults based on each rule, NNmf is used to expressWhen multiple detection methods are used in thethe membership of each rule for input X andfuzzy inference system, each detection method can aggregate each result. The output of the system is aslead to one consequence at first. Then a conclusion follows中国煤化工can be aggregated by each consequence throughTHgCNMHGfuzzy reasoningJ. China Univ of Mining Tech( English EditionVol 16 No. 392 pairs of data for experiment are collectedfrom the working face, laneway, nearby tube of thewell. The sample data collected are divided into twoparts. 56 pairs of data are used for training. Theremainder 36 pairs of data are used to evaluate theperformance of the system. And the identificationresults are contrasted with raw data(Fig. 2)Of the 36 pairs of data, 32 pairs are inaccordance with the raw data. About 90%o reliabilitycan be offered by the fuzzy system.Fig. I Fuzzy system based on neural networkswhere 8,-8 are the results under single detectionmethod. u,-uk the outputs of NNmf, each output isa value with a range of (0, 1) represents themembership function of g3.2 Structure of the fuzzy neural network04The 3-layer bP network is selected for eachResultsneural network. Each input node of NNi representslinguistic term of an attribute 2.For instanceRaw data ofactual instancetemperature is an attribute; it can be described byusing three linguistic terms(high, medium and low)1015202530354Fuzzification has completed before the detection databeing sent to neural network, the final input value of Fig. 2 The contrast of actual instance and experiment resultNNi is a vector of the membership degree. Themembership degree indicates to what degree the 5 Conclusionsalue of attribute belongs the linguistic term. Thenumber of linguistic terms of each attribute andThe characteristics of spontaneous combustionmembership function is decided by actual instance and the complexity of its detection are introducedand experiences. The output of NN represented by first in the paper. A fuzzy inference system based onthe membership of combustion probability is theartificial neural network and CRI is put forward todetection result under single detection methoddetecting the spontaneous combustion. Someconclusions can be gotten as fol4 ExperimentBefore recognizing and detecting the coalspontaneous combustion, the condition of coalminesGuqiao coal mine in Huainan mining area was and the characteristics of the detection methodsselected as the experiment spot. Thermometry and should be analyzed. These are necessary to establishdiometry are adopted as two detection methods in the inference rulesthe experiment. Thus only three BP neural networksMultiple detection methods are adopted for theirare adopted in the systemcomplementarities in detecting combustion. The CriEach pair of raw data includes input and output reasoning methods and FITA are suitable for multi-data to represent the system input and outputrules in the environment of detecting coalminerespectively. The system input includes the spontaneous combustiontemperature and the CO value of detection spot whilThe combination of the neural network andthe system output includes probability of spontaneous fuzzy system has mutual benefit function. Thecombustion onlyeffectiveness of the system is proved by experiment( Continued on page 265)中国煤化工CNMHGWANG Xiao-hong et al.Research on Surface Modification of 96 Al-O, by Ni ion implantationReferences[1 Wang H, Zeng L K, Wu J Q, et al. Present status and prospect of surface modification of ceramics Materials Protection, 2000,33(11):43-46.( n Chinese)[2 Wang Q Z, Chen Y F. The surface modification of ceramics by ion implantation and ion beam mixing Nuclear Techniques1994,17(9):569-574.( n chinese)[3 Gong T, Deng K, Zhao B R, et al. Improvement of mechanical properties of Al2O3 ceramics through Ni ion implantation[4] Xu T, Yang SR, Lu J J, et al. The influence of ion implantation on structure and anti-wear ability of single crystal SiOJournal of Chinese electron Microsope Society, 2000, 19(4): 597-598.(In Chinese)[5] Xiang G, Wang C, Zheng S K, et al. Improve the photocatalytic activity of nanometer TiO2 thin film by ions implantationJournal of Functional Materials and Devices, 2002,8(1): 24-27. (In Chinese[6 Ma X M, Hu J B, Shang J, et al. Electrochemical behavior of coenzyme I and its application on glassy carbon ion implantationmodified electrode. Chinese Journal of Analytical Chemistry, 2004, 32(8): 1027-1030 (In Chinese)[7 Li C. Epoxidation of olefins on catal ysts with highly isolated transition metal ions. Chinese journal of Catalysis, 2001, 22(5)479-484.(In Chinese)[8 Bhansali S, Sood D K. Selective seeding of copper films on polyimide-patterned silicon substrate, using ion implantationSensors and Actuators, 1996, A52(1-3): 126-131[9] Bhansali S, Sood D K. A novel technique for fabrication of metallic structures on polyimide by selective electroless copperplating using ion implantation. Thin Solid Films, 1995, 270(1-2): 489-492[10 Li J H, Tsai Y Y, Chiu SY, et al. Palladium seeding on the tantalum-insulated silicon oxide film by plasma immersion ionmplantation for the growth of electroless copper. Thin Solid Films, 2000, 377-378: 592-596[11 Wang X H. Study on the Technique of Ni lon Implantation Assisted Electroless Copper Plating on Surface of A 2O3 Ceramics[Master's degree dissertation]. Xuzhou: China University of Mining and Technology, 2004.(In Chinese)[12] Radermacher K, Mantl S, Dieker C, et al. Ion beam synthesis of buried alpha-FeSi2 and beta-FeSi2 layers. Applied PhysicsLetters,1991,59(17):2145-2147.[13] Ila D, Williams E K, Sarkisov S, et al. Formation of metallic nanoclusters in silica by ion implantation. Nuclear InstrumentsInd Methods in Physics Research section B, 1998, 141(1-4): 289-293[14 Wei Z s, Wang M, Zuo D W, et al. Changes of surface roughness of TC4 alloy implanted by La, Mo. Special CastingNonferrous Alloys, 2003(4): 1-3 (In Chinese)[15 Carlsen T A Phoelectron and Auger Spectromete. Beijing: Science Press, 1983.(In Chinese)[16 Zhang T H, Wu Y G. The Surface Optimization Technology of lon Implation. Beijing: Metallurgical Industry Press, 1993.(InChinese)(Continued from page 260)References[1] Sun J P, Song S. Recognition of spontaneous combustion in coal mine based on genetic clustering. Journal of chinaUniversity of Mining &Technology( English Edition), 2006, 16(1): 42-43[2] Sun J P, Song S Image recognition of coal mine spontaneous combustion and synthetic criterion. China Safety ScienceJournal,2005,15(12):110-112.( n Chinese)ZAdeh L A Fuzzy sets. Information and Control, 1965(8): 338-3534] Li F Fuzzy Expert System. Wuhan: Huazhong University of Science Technology Press, 1995.(In Chinese[5] Zadeh L A. a theory of approximate reasoning. Machine Intelligence 1979, 9: 149-194[6 Zadeh L A. Outline of a new approach to the analysis of complete system and decision process. IEEE Transactions on SystemsMan Cybernet, 1973, 3: 2844[7] Bandler M, Cohout L J. Fuzzy power sets and fuzzy implication operator. Fuzzy Sets and Systems, 1980, 4: 13-30ou c Man, and Cybernetics, Oct. Tucson: IEEE, 2001: 2145-2150v reasoning 200/ IEEE International Conference on[9] Zhou C G, Liang YC, Yang Z M. A fuzzy neural network based on fuzzy weighted reasoning method. Proceedings ofInternational Workshop on Autonomous Decentralized SystemS, 2000: 190-195[10 HollandJ H, Adaptation in Natural and Artificial Systems. Ann Arbor: University of Michigan Press, 1975[Il Zhao ZY, Xu Y M. Introduction to Fuzzy Theory and Neural Network and Their Application. Tsinghua University Press,Guangxi Science& Technology Publishing House, 1995.(In Chinese)[12] Fang T G, Wang XZ. A new approach to weighted fuzzy production rule extraction from neural network, Proceedings of theThird International Conference on Machine learning and Cybernetics. Shangh中国煤化工CNMHG