Acoustical　signals　from　mechanical　systems　reveal　the　operational　status　of　mechanical　components,　which　can　be　used　for　machinery　condition　monitoring　and　fault　diagnosis.　However,　it　is　very　difficult　to　extract　or　identify　the　acoustical　source　features　as　the　measured　acoustical　signals　are　mixed　signals　of　all　the　sources.　Therefore,　this　paper　studies　on　the　source　separation　and　identification　of　acoustical　signals　using　principal　component　analysis　and　correlation　analysis.　The　effectiveness　of　the　presented　method　is　validated　through　a　numerical　case　study　and　an　experimental　study　on　a　test　bed　with　shell　structures.　This　study　can　provide　pure　acoustical　source　information　of　mechanical　systems,　and　benefit　for　machinery　condition　monitoring　and　fault　diagnosis.