Control and Monitoring of Manufacturing Processes
Industries are constantly seeking efficient tools to control and monitor increasingly complicated manufacturing procedures, and thereby relying extra on computer integrated manufacturing or CIM and automation. The human operator's success in process control and monitoring tasks suggests as one probable approach to implementing computer based control and monitoring systems is to model the human operator's learning and decision making capabilities. An intelligent controller should posse's capabilities to learn from illustrations and employ knowledge gained throughout a learning process to optimize the operation of machines. Such is analogous to the procedure of learning by that a novice human machinist as an expert. Neural networks such are promising tools for online monitoring of difficult manufacturing process. Their better learning and fault tolerance abilities enable high success rates for controlling or monitoring the machining processes. In between the manufacturing applications of neural networks, controlling and monitoring can be taken in two dimensions as: the control and monitoring of work pieces that is surface finish, machines and automatic setups for example: vibration, thermal deflection and tool wear. Neural networks are trained via examples, hence eliminating require for explicit mathematical modelling.
The feed-forward neural networks are utilized to recognize happens of tool wear in turning operations. The neural network trained along with the BP algorithm is learned to perform tool-wear detection described information from the sensors on cutting force and acoustic emission. The better learning and fault tolerant abilities of the neural network donate to the high success rates in recognition of tool wear. Conversely, design parameters as like: training parameters, sensors, and network structure utilized affect the performance of the system. A multilayer feed-forward neural network has been applied to predict the improvement of chip break-ability and surface terminates in a machining process at various tool wear states. In the primary phase, chip forming patterns that is chip breaking or shapes were estimated beneath the condition of an unworn tool. Subsequently, the neural networks were trained along with input features as like: dispersion patterns, and initial prediction cutting, parameters of break-ability and output in terms if fuzzy membership value of chip break-ability and surface roughness. Subsequent to offline training utilizing the BP algorithm, the neural network was capable to successfully predict online machining performance as like: chip break-ability, surface finish, tool wear and chip shapes. The neural network is able of predicting chip forming patterns offline and also updating them on line as tool wear improves. This technique is applicable for any type of tool configuration, or and rough machining situations. Moreover, neural networks have been utilized for minimizing thermal deflection and a neural network stand on thermal spindle error compensation system is being utilized. The temperatures on the positions of a milling machine were controlled and monitored and fed into a multilayer feed-forward neural network trained by the Back Propagation algorithm to predict the thermal deflection of three principal spindles. The evaluated thermal errors were adopted by the CNC controller that send out the control signals are compensated to drive the milling machine. The neural network illustrated the prediction accuracy of more than 85 percent in varying and new cutting situations.
The hierarchy of neural network application for process control and monitoring by the present applications comprises, machining process monitoring, tool wear monitoring, process control and also process modeling. The neural network models utilized were multilayer feed-forward networks, ART networks, and cerebellar model articulation controller. For the online monitoring of complex processes, and react as appropriate modeling tools in conditions lacking several information, or where analytical modeling would be more complex, neural networks are promising tools. Additionally, their fault tolerant and advanced learning abilities enable high success rates to monitoring machining processes. The adaptive capacity of neural networks additionally make them excellent candidate for control and monitoring applications. A neural network monitor react as one of the mainly efficient tools in determining the appropriate set of manufacturing parameters via predicting the consequence of machining parameters to the machining process earlier. Neural network's applications appear also promising for real or genuine time no-linear mapping of distorted input data vectors. Identification of techniques as like a package of tools such could be combined in an exacting application may be the main key to future intelligent control. Analysis of systems incorporating neural network into real time control systems should allow the latter to optimize the presentation on line utilizing variables that or else would require sophisticated models, algorithms, and complex computation. The parallel computation capabilities of neural network present the possibility for developing intelligent systems such are capable to learn from illustrations, in real time manufacturing environment, recognize process patterns, and initiate control actions.