Scientific discovery: computational explorations of the creative process
Scientific discovery: computational explorations of the creative process
C4.5: programs for machine learning
C4.5: programs for machine learning
Rough Sets: Theoretical Aspects of Reasoning about Data
Rough Sets: Theoretical Aspects of Reasoning about Data
Machine Learning
| Hi-index | 0.00 |
The main task of machinery diagnostics consists of inferring the actual state of a given machine based upon observed symptoms and operating conditions. Diagnostic knowledge is commonly presented in a declarative form. Precise predictions require functional dependencies. Such knowledge can be elicited from data collected during operation of the machine or generated using simulation software, the latter case being discussed in this case study. Several independent attributes representing the operating conditions and technical state of a rotating machine were varied systematically, causing the response--vibrations at several points of the machine. To acquire knowledge from data we used several machine learning techniques and 49er, a system that can discover various forms of knowledge. First we focused the search on detection of groups of attributes that approximate functional relations between control and dependent variables. Then Equation Finder was applied recursively to each group, finding multidimensional equations that capture fine functional relationships. Since diagnostic inference takes place in the direction opposite to causal relationships, the discovered equations had to be inverted, so that the observed vibrations of the machine enabled us to calculate values describing the internal state of the object. The knowledge discovered using this approach was implemented in a diagnostic expert system.