Methods for intelligent online monitoring and diagnosis of manufacturing processes.

The study of on-line monitoring and diagnostics of manufacturing processes can be classified into four aspects: manufacturing processes, monitoring, diagnosis, and computer implementation. This thesis focuses on monitoring and diagnosis and develops four new methods: a monitoring method, a monitoring alarm threshold estimation method, and two diagnostic methods. The monitoring method is developed based on the dynamic data system (DDS) methodology. It uses the eigenvalue locations of the process auto-regressive moving average (ARMA) model and is equivalent to the damping ratio method but reduces the computational load significantly. This method is successfully applied in the tool wear monitoring of a milling process. The alarm threshold estimation method is a heuristic method developed based on two assumptions: (1) the given monitoring index can recognize the process conditions at any process working condition, and (2) the normalized variations of the index are about the same at different process working conditions. This method is tested in a spindle imbalance monitoring application. The key for the diagnosis is the relationship between process conditions and diagnostic indices. The first diagnostic method uses the fuzzy set theory to build a linear fuzzy equation to describe the relationship. By solving the fuzzy equation, a diagnostic decision can then be made. To demonstrate this method, an application in diagnosis of a tapping process is presented. The second diagnostic method originates from Artificial Intelligence (AI) and describes the relationship by a decision tree. It consists of three parts. First, a decision tree is built by applying the machine learning algorithm ID3. Second, the tree is pruned by the proposed tree modification method. Finally, a decision is made based on a new decision-making strategy. This method is applied to the same tapping process and to a welding process. This thesis also describes two computer implementations: (1) A multi-modular processors system (MMPS), which is a hierarchical computer system designed for on-line monitoring of a super die manufacturing cell. (2) A single-board computer implementation for on-line ARMA model parameter estimation.Ph.D.Mechanical engineeringComputer scienceArtificial intelligenceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/162301/1/9001621.pd