Knowledge-based systems using neural networks for electron beam welding process of reactive material (Zircaloy-4)

Bead-on-plate welding of zircaloy-4 (a reactive material) plates was conducted using electron beam according to central composite design of experiments. Its predictive models were developed in the form of knowledge-based systems in both forward and reverse directions using neural networks. Input parameters considered for this welding of reactive metals were accelerating voltage, beam current and weld speed. The responses of the welding process were measured in terms of bead width, depth of penetration and micro-hardness. Forward mapping of the welding process was conducted using regression analysis, back-propagation neural network (BPNN), genetic algorithm-tuned neural network (GANN) and particle swarm optimization algorithm-tuned neural network (PSONN). Reverse mapping of this process was also carried out using the BPNN, GANN and PSONN-based approaches. Neural network-based approaches could model this welding process of reactive material in both forward and reverse directions efficiently, which is required for the automation of the same. The performance of the neural network models was found to be data-dependent. The BPNN could outperform the other two approaches for most of the cases but not all in both the forward and reverse mappings.     

Multilevel segmentation of Hippocampus images using global steered quantum inspired firefly algorithm

Applied Intelligence - Microscopic Image segmentation has a crucial role in detecting and diagnosing numerous critical diseases like Alzheimer’s disease, Kidney disease, Cancer, many...     

Near-optimal gait generations of a two-legged robot on rough terrains using soft computing

A two-legged robot will have to generate its near-optimal gaits after ensuring maximum dynamic balance margin and minimum power consumption, while moving on the rough terrains containing some staircases and sloping surfaces. Moreover, the changes of joint torques should lie below a pre-specified small value to ensure its smooth walking. The balance of the robot and its power consumption are also dependent on hip trajectory and position of the masses on various limbs. Both neural network- and fuzzy logic-based gait planners have been developed for the same, the training of which are provided using a genetic algorithm off-line. Once optimized, the planners are found to generate optimal gaits of the two-legged robot successfully for the test cases.     

A comparative study on some navigation schemes of a real robot tackling moving obstacles

A comparative study of various robot motion planning schemes has been made in the present study. Two soft computing (SC)-based approaches, namely genetic-fuzzy and genetic-neural systems and a conventional potential field method (PFM) have been developed for this purpose. Training to the SC-based approaches is given off-line and the performance of the optimal motion planner has been tested on a real robot. Results of the SC-based motion planners have been compared between themselves and with those of the conventional PFM. Both the SC-based approaches are found to perform better than the PFM in terms of traveling time taken by the robot. Moreover, the performance of fuzzy logic-based motion planner is seen to be comparable with that of neural network-based motion planner. Comparisons among all these three motion planning schemes have been made in terms of robustness, adaptability, goal reaching capability and repeatability. Both the SC-based approaches are found to be more adaptive and robust compared to the PFM. It may be due to the fact that there is no in-built learning module in the PFM and consequently, it is unable to plan the velocity of the robot properly.     

Soft computing-based approaches to predict energy consumption and stability margin of six-legged robots moving on gradient terrains

Soft computing-based approaches have been developed to predict specific energy consumption and stability margin of a six-legged robot ascending and descending some gradient terrains. Three different neuro-fuzzy and one neural network-based approaches have been developed. The performances of these approaches are compared among themselves, through computer simulations. Genetic algorithm-tuned multiple adaptive neuro-fuzzy inference system is found to perform better than other three approaches for predicting both the outputs. This could be due to a more exhaustive search carried out by the genetic algorithm in comparison with back-propagation algorithm and the use of two separate adaptive neuro-fuzzy inference systems for two different outputs. A designer may use the developed soft computing-based approaches in order to predict specific energy consumption and stability margin of the robot for a set of input parameters, beforehand.     

Effect of powder treatment on the crystallization behaviour and phase evolution of Al2O3–High ZrO2 nanocomposites

Al₂O₃–ZrO₂ composites containing nominally equal volume fraction of Al₂O₃ and ZrO₂ have been synthesized through combined gel-precipitation technique. Subsequently the gels were subjected to three different post gel processing treatments like ultrasonication, ultrasonication followed by water washing and ultrasonication followed by alcohol washing. It was observed that while in unwashed samples crystallization took place at low temperature, crystallization was delayed in the washed gels. The phase transition of ZrO₂ in the calcined gels followed the sequence; amorphous → cubic ZrO₂ → tetragonal ZrO₂ → monoclinic ZrO₂. On the other hand, phase transition in alumina followed the sequence amorphous to γ-Al₂O₃, the transition taking place at 650 °C. No α-Al₂O₃ could be detected even after calcination at 950 °C. However, all the sintered samples had α-Al₂O₃. In spite of high linear shrinkage (19–21%) during sintering, the sintered sample had density of only above 70% for all the four varieties of the powders. However, in spite of the low sintered density of the pellets, 31% tetragonal zirconia could be retained after sintering at 1400 °C and it reduced to about 16% at 1600 °C.     

Dielectric and Thermal Investigations on PbNb2O6 in Pure Piezoelectric Phase and Pure Non-Piezoelectric Phase

A high piezoelectric transition temperature is indicated for present orthorhombic samples of lead meta-niobate (PbNb₂O₆) by the high value of the temperature (T_m) at which the real part (?′) of electrical permittivity peaks. It is 573 to 580°C. The imaginary component (?″) increases sharply at these temperatures. Cooling run after heating to 700°C gives changed values of ?′ and ?″ and T_m. Present work reports impedance spectroscopy (up to 700°C over 20 Hz to 5.5 MHz) and Differential Scanning Calorimetry or DSC on the pure orthorhombic phase, and, for the first time, on the pure rhombohedral phase. DSC shows an endothermic dip across T_m during heating and an exothermic peak during cooling for the orthorhombic sample only.     

Modeling of plasma spray coating process using statistical regression analysis

This study aimed to correlate outputs of a plasma spray coating process with its input parameters. Central composite design of experiments had been followed to conduct the experiments on plasma spray coating process, in order to collect its input–output data. The effects of four input parameters, namely primary gas flow rate, stand-off distance, powder flow rate, and arc current on three outputs (responses), such as thickness, porosity, and microhardness of the coatings, had been studied using non-linear statistical regression analysis. Such an analysis yielded the response equations, and viability of these equations had been tested on a set of experimentally obtained cases. The above four process parameters were found to have varying influences on different responses. However, all the above three responses were found to be significantly dependent on primary gas flow rate.     

Optimization of bead geometry in electron beam welding using a Genetic Algorithm

Bead-on-plate welds were carried out on austenitic stainless steel plates using an electron beam welding machine. Experimental data were collected as per central composite design and regression analysis was conducted to establish input–output relationships of the process. An attempt was made to minimize the weldment area, after satisfying the condition of maximum bead penetration. Thus, it was posed as a constrained optimization problem and solved utilizing a Genetic Algorithm with a penalty function approach. The Genetic Algorithm was able to determine optimal weld-bead geometry and recommend the necessary process parameters for the same.