Vishy Karri

Fuzzy Expert System to Estimate Ignition Timing for Hydrogen Car

This paper presents the application of fuzzy expert system technique as a basis to estimate ignition timing for subsequent tuning of a Toyota Corolla 4 cylinder, 1.8l hydrogen powered car. Ignition timing prediction is a typical problem to which decision support fuzzy system can be used. Based on extensive experiments, the basic fuzzy rules on ignition timing have been constructed, in which the engine speed, throttle position, manifold air pressure, fuel pulse width, engine power, lambda value were chosen as fuzzy sets of the linguistic input variables, and ignition advance is selected as performance output of the fuzzy system. The constructed fuzzy system initially mapped 136 basic rules based on physical theories and extensive experimentation. For all the input parameters various triangular, trapezoidal and generalized bell-shaped membership functions were successfully applied to best represent the ignition timing output from the expert system. The results have shown that the minimum ignition advance for maximum torque without detonation was achieved. The estimation of ignition advance achieved from fuzzy expert system was ± 5% root mean square error.

Tool condition monitoring in drilling using artificial neural networks

A reliable and sensitive technique for monitoring tool condition in drilling is essential help for practising engineers. It is commonly known that the unattended use of a drill bit until it reaches the ultimate failure can potentially damage to machine tool and work-piece resulting in considerable down time and productivity loss. Thus there is a need for such tools to save high maintenance costs in case of the catastrophic failure. A system in drilling that can estimate tool life in terms of the number of hole to failure as condition monitoring techniques in the form of a digital display is significantly beneficial. In this paper, a tailor-made novel feed forward network is proposed to predict tool life in terms of the number of holes to failure. These involved the development of predictive model, test rig design and a digital display to assist engineers with on-line tool life. To entitle the network to cater for various cutting conditions, a knowledge base as training and testing data have to be generated on the experimental data in a comprehensive working range of drilling. Consequently, the experiments were performed in thirty-two cutting conditions based on the combination of three basic cutting parameters, which are feed rate, spindle speed and drill diameter. The neural networks were trained and the architecture of networks was appropriately selected by benchmarking the Root Mean Square error (RMS). The results of the novel network, Optimisation layer by layer (OLL), have shown the ability to accurately predict the number of holes to failure with a 100% success rate at both training and testing stages. To highlight OLL predictive capability, a brief comparison with Backpropagation Neural Network (BPNN) is carried out.

A Monitoring System of Drill Wear States Using a Hybrid Neural Network

A reliable and sensitive technique of cutting tool condition monitoring in drilling is essential help for practising engineers. It is commonly known that a worn drill bit produces a poor quality hole. In extreme cases, a catastrophic failure of a drill bit during cutting can destroy a work-piece and damage a machine tool resulting in low productivity and expensive down time To detect the states of cutting tool wear condition, attempts are made to physically measure wear land. An intelligent system for detecting wear condition without interrupting the process is essential to avoid unexpected cutting tool breakage. In this work, an intelligent algorithm is proposed to real-time monitoring of drill wear states, in the form of a digital display, over a comprehensive range of cutting conditions. A novel neural network, Hybrid Neural Network (HNN), was developed and tested in this task. The results of the HNN have shown the ability to accurately monitor the wear states up to a 92% success rate. With such a highly accuracy of results, the developed system can be used for monitoring wear states in drilling and warning operators.

Race car chassis tuning using artificial neural networks

Proficient chassis tuning is critical to the overall performance of any race car. Determination of the optimum arrangement for specific track conditions can require a large amount of practical testing and, as such, any tools that reduce this expenditure will be of great value to the racing industry. Traditional computer modeling based on simplified vehicle dynamics has had a growing use in this field, but due to the extremely complex nature of the vehicle / driver / environment entity it has a number of practical limitations. Intelligent models, such as Artificial Neural Networks, on the other hand are not limited in this way and show a number of potential benefits. This study presents a simplified application of ANN to predict the optimum chassis arrangement for a steady state cornering condition for a Formula SAE race car to see if these benefits can be realised. The race car was equipped with a large sensor array, including engine speed, throttle angle, wheel speed, suspension position, steering angle, longitudinal and lateral acceleration and yaw rate, and chassis tuning was accomplished by varying caster, toe and front and rear tyre pressures. Data was collected for a total of six different chassis tuning combinations for the steady state cornering condition and a feed-forward back-propagation ANN model capable of predicting the lateral (centrifugal) acceleration of the vehicle for any given chassis tuning was produced. A numerical investigation was then completed with the ANN model to find the maximum lateral acceleration, and therefore speed, of the vehicle for each different possible chassis tuning combination. Each of the resulting 480 combinations were then ranked and compared against the optimal combination found from extensive practical vehicle testing. Despite a few problems encountered throughout the investigation that deteriorated ANN model accuracy, a high degree of correlation was found.