Algorithm for Controlling Mechanical Properties of Hot Rolled Steels Using Bayesian Network Model
The work presents a control method for on-line adjustment of mechanical properties of as rolled steels produced at hot strip mills. The key idea of the method is a probabilistic causal (Bayesian) network which represents in a form of a directed acyclic graph the joint probability distribution of mechanical properties, chemical composition and temperature–strain parameters acting during hot rolling. As a slab moves along the mill the distribution is used for continuous recalculating the posterior probability of all mechanical properties conditioned by chemical composition and all other process parameters which become known to the moment of recalculating. Finally, when a strip is just before the finishing group we evaluate the probability distribution of finishing rolling temperature and coil temperature given the strip has the target mechanical properties. It generates new setups for these temperatures A pilot version of the method has been just implemented at CSP–line at Vyksa, Russia, United Metallurgical Company’s steel production site The adjustment is realized through appropriate correction of finish rolling temperature or/and coiling temperature setups of the mill automatic control system after the last chemical analysis of the current heat is made at the start of casting. Only “cautious” corrections of the temperatures are permissible so far (deviation from predefined level not more than ±30 degrees for each temperature) and the main aim of them is to set off the influence of chemistry variations on mechanical properties scatter of a given steel grade. The results of using the algorithm show that even these limited but interconnected actions reduce approximately twice the standard deviation of the mechanical properties inside a steel grade.