An Intelligent Ae Sensor for the Monitoring of Finish Machining Process

The paper presents the latest results of sensing the cutting process on the basis of AE signals and some particularities in further development of the monitoring model for the finish turning process. Due to non-linearity, the large number of influencing parameters and missing information in AE data, the Artificial Neural Networks were chosen as a monitoring decision tool. The problem of accurateness in predicting the surface roughness on the basis of AE because of the mutual interdependence of the data requires a special procedure for building a neural network model. The final aim of such an approach is presented as improvements in learning or considerable reduction in error prediction. Further development of the monitoring model has the goal of building a so-called intelligent sensor, which should be able to perform the signal conditioning and feature extraction process. INTRODUCTION Most of the reports on research into the machining processes usually start with a similar ascertainment: the complexity of the cutting process is one of the main obstacles to successful modeling or monitoring of processes; this fact gives us the impetus to continue permanent investigations. There are no simple answers or quick solutions; a reliable monitoring approach or a complete control system for the cutting process is a task in which successful solutions could be obtained only through numerous, systematic investigations covering the different scientific areas incorporating sensor technologies, signal processing techniques, modeling methods, etc. Probably one of the best of the latest reviews of such efforts has been made within the CIRP groups, where the conclusions stated that the different monitoring systems with acceptable commercial reliability are now available in the market, although the narrow range of performance provides only limited applicability (Byrne and others at [1]). The report also confirms one of the main gaps in this kind of research; i.e. to develop a system as an integrated part of an intelligent machine tool, much more should be done at both the hardware and software level to obtain a simple and reliable sensor for machining applications. At present when the development of manufacturing processes heavily depends on information technologies the realistic process models are also one of the prerequisites for predicting the performance of metal cutting operations. The latest report on the modeling of machining operations (C.A. van Luttervelt and others at [2]) concluded, that most of the research deals with possible new ways of obtaining better control of machining operations; however a common framework is still missing. At the first IPMM conference the monitoring concept on the basis of sensing Acoustic Emission signals in finish machining processes was presented (see Dolinsek at [3]). From the contents of the AE signals we were able to extract significant features from the process, depending on the cutting conditions, which serve as learning data for the ANN structure. The model should be applicable for practical cutting in such a way that the predicted values of surface roughness could be a sign to adapt the cutting parameters in order to achieve the required surface quality or to detect disturbances in the process (tool wear, unfavorable chip shape, lack of coolant). In the introduction we also draw attention to the lack of adequate sensors and indicate that the sensing technology will play an important role in the development of future manufacturing systems. Further investigation of our monitoring concept for finish machining processes was therefore oriented towards the search for reliable sensing. Some of the results using the AE-jet sensor were discussed at SEM and CIRP conferences (see Dolinsek at [4] and [5]). The main advantages of this sensor were presented as improvements of the signal to noise ratio, simple upgrade, and the fact that the cutting process and sensor are not reciprocally disturbed. Through the spectral analysis technique, and with adequate averaging procedures, we were therefore able to gain some useful information for the further development of our monitoring model. Artificial neural networks (ANNs) were used as a operating tool because they can handle strong non-linearites, a large number of parameters, missing information, and the characteristics of the data which are also significant in our monitoring approach. Based on their inherent learning capabilities, ANNs can adapt themselves to changes in the production environment, and can also be used in case where no detailed information is available about the relationships among the various manufacturing parameters. In many cases, there is also no exact knowledge about the relationships among parameters; it is unknown which input-output configuration of an ANN can satisfy the accuracy requirements. Therefore, a method is needed for automatic input-output configuration of the applied ANN model. This paper therefore addresses the problem of automatic input-output configuration and generation of ANN-based monitoring models, i.e. those parameters to be considered as inputs, and those as output, in order to accurately predict surface roughness and classify tool wear in the finish turning process. AE-JET SENSOR FOR MONITORING A FINISH MACHINING PROCESS In researching Tool Condition Monitoring (TCM) systems for the manufacturing processes and introducing them to the workshop environment we are engaged in solving three main tasks : • building up a sensor system which is reliable for sensing the process parameters with minimal influence on the process, • applying proper signal processing techniques capable of processing the real life signals, • developing decision-making algorithms capable of estimating the process conditions. In such an feature-based approach, we observe some features, extracted from sensor signals in order to identify different process conditions and compare them to normal and unfavorable cutting conditions. This process is generally not too complicated, but the success of the monitoring depends greatly on exact correlation of the measured parameters to the cutting process characteristics i.e. the sensors are the first and main component leading to the successful solution of our tasks. Once we find or build-up a sensor which satisfies the main requirements demanded in practical monitoring approaches: measurement close to the machining point; no influences on the machine-tool characteristics; function independent of tool or workpiece; low costs; maintenance and wear free; resistance to dirt and to mechanical and thermal influences, minimal reciprocal disturbances between the process and sensor, simple upgrade which allows easy further improvements. Thus we can further develop our monitoring model by applying signal processing, feature extraction and decision making procedures. When this intelligent part is successfully solved the final hardware integration into the sensor is not a complicated task. One of the most promising tool monitoring techniques is based on sensing the Acoustic Emission (AE) signals generated at the cutting zone. Extensive publications have demonstrated the extreme sensitivity of AE signals to certain process parameters (etc. Dornfeld at [6]). In general it is agreed that during metal cutting, plastic deformation (continuous type of AE signals) and fracture of the material (burst type of AE signals) are major sources for AE waves. One of the basic researches of the AE phenomena in the cutting is that made by Moriwaki, who illustrated in detail seven possible sources of the AE signals in the vicinity of the cutting process [7]. However in sensing and analyzing the AE signals generated in the cutting process we will always face two main obstacles: • it is almost impossible to built-up a physical model of the AE signal in relation to the AE waves, since the signal generated in real cutting processes in complex workpiece structures is continuous and random, • we expect from the sensor used in metal cutting problems a reliable sensing of AE signals from all sources of generation of the signal with the ability to differentiate between the particular sources, but without any other interference’s. Due to those limitations the most common approach of the application of the AE in monitoring of cutting processes is at present still a simplified sensing of the mixed AE signals. From the content of the acquired AE signals using a suitable postprocessing procedure one can then identify different process conditions. Although many different sensors are available for AE measurements, only few can be used in a machine tool in which aggressive ambient conditions occur. The main disadvantages of traducers, which are mainly designed for non-destructive inspection or research work, are that they cannot withstand the high temperatures, large coolant volumes and abrasive wear through chips. With a new concept of AE transducers (see [8]), the Water-jet AE sensors a liquid or coolant stream is used as a transmission medium to transfer AE signals generated from the cutting process to the PZT element. As the distance between the cutting zone and transducer element is small, the damping effect is minimized, and the signal-to-noise ratio is significantly improved. The construction of the sensor, developed for our monitoring task, presented in Fig 1, was a practically built-in CNC finish turning machine. The applicability of this sensor in the finish turning process was tested throughout the proper analysis of the acquired AE signals and further relation of their content to the process conditions.