In　order　to　investigate　the　source　and　chemical　composition　of　precipitation　were　meased　at　eight　sites　in　the　pusan region　during　the　period　from　February　to　September 1995．289 samples　of　precipitation　were　collected　by　automatic seperation　sampler　of 6－step，and　the　amount　of　dissolved　components（H⁺，Na^＋，K^＋，NH₄^＋，Ca^2＋, Mg^2＋, C1^-，SO₄^-, NO₃^-）and　conductivity　were　measured　by　ion　chromatography，pH　meter　and　conductivity　meter． In　this　study，the　avefage　of　total 289 samples　pH　and　electric　conductivity　xvere 5.99 and 70.3μs／cm，and　according to　average　concentrations　of　components　were　composed　of　SO₄^2-，Ca^2＋，C1^-，NH₄^＋，Na⁺ Mg^2＋，NO₃^-，K^＋ with 221， 159.8，157，99.9，96.2，56.5，50.9 and 24.6μeq／1，respectively， Colltribution　of　non－seasalts　were　larger　than　the　seasalt，which　has 71.3 ％ and　among　the　rest 81.3％ was　neutralized by　ammonia　and　calcium　species　to　acidfy　compounds，such　as　H₂SO₄ and　HNO₃ in　the　procipitation． Especially，source　of　calcium　compounds　was　estimated　from　soil，so，it　seems　that　precipitation　is　neutralized　by　Ca²⁺ than　NH₄⁺in　pusan，korea.

Neural　networks　have　recently　become　the　focus　of　much　attelltiQn，largely　because　of　their　wide　range　of　complex　and nonlinear　problems．This　paper　presents　a　new　integrated　approach　using　neural　networks　for　diagnosing　process　failures as　an　example　of　amlysis　for　a　nitric　acid　cooler　process．The　fault　propagation　in　process　is　mQdeled　by￠ausal　relation・ships　from　the　fault　tree　and　its　minimal　cut　sets．The　measurement　pattems　required　for　training　and　testing　the　neural networks　were　obtained　from　fault　propagation　model．The　fault　diagnostic　multi　neural　network　has　the　hierarchical structure　in　which　the　circuit　level　network　and　the　component　le▽el　networks．The　circuit　level　network　leams　malfmc－ tions　of　control　loops（feedback　and　feedforward　type　control　circuits　and　protective　systems）．The　component　level networks　leam　component　failures　and　disturbances．The　measurement　pattems　relative　to　malfunctions　of　control　loops are　obtained　from　top　structures　of　fault　trees　for　process．The　measurement　pattems　relative　to　component　failure　and disturbances　are　obtained　from　a　part　of　the　fault　trees　for　the　control　loops．The　circuit　level　network　first　can　identify malfunctions　of　control　loops　by　matching　the　pattem　of　process　data　from　plant　with　the　pattem　of　training　data　which network　leamed．Secondly，the　component　Ievel　networks　can　identify　component　failures　and　disturbances　which　can　lead to　system　malfunctions．