Sheshadri, AK and Jangid, P and Singh, S and Bhargaw, HN and Botre, BA and Akbar, SA and Hasmi, SAR
(2019)
AI Models for Prediction of Displacement and Temperature
in Shape Memory Alloy (SMA) Wire.
In: 4th International Conference on Emerging Technologies: Micro to Nano (ETMN-2019), December 16-17, 2019, Pune, India.
Abstract
Shape Memory, Alloys (SMAs) are a unique class of smart materials that have the ability to recover their shape on temperature stimuli_ During this transformation, hysteresis and non-linear behaviour can be observed and open-loop control design is inadequate for tracking control of these actuators. This is a major setback for the design and development of any SMA device. The main hurdle is that this nonlinearity can't be modeled effectively even by 3rd-degree differential equations. Additionally, the apparatus used for measurement of strain recovery during SMA actuation includes linear variable differential transducer (LVDT), which is very bulky and expensive and doesn't let us fully utilize the potential of SMA applications in miniaturized devices. This research work presents a method to eliminate the bulky position sensor by introducing an Artificial Neural Networks (ANN) to compensate for the non-linearity. Various researchers have attempted to model the behaviour of SMAs using ANN techniques but these models had a high RMS error. In this paper. we develop a more complex neural network to model SMA's behaviour. We model SMA's behaviour with (i) displacement prediction (ii) temperature .prediction. The results of our research not only demonstrate the effectiveness of Neural Networks for prediction of displacement and temperature of the SMA but also show how the proposed architectures have a much lesser error as compared to earlier models and are much more effective at modeling SMAs.
Shape Memory Alloys (SMAs) are a unique class of smart materials that have the ability to recover their shape on temperature stimuli. This Shape Memory Effect (SME) occurs due to temperature stimuli and stress-dependent shifts which causes a change in the material's crystalline structure between two different phases - martensite (low-temperature phase) and austenite (high-temperature phase). During the phase transformation, hysteresis and non-linear behaviour can he observed which are a major setback for any device design and development. Therefore, most SMA applications are based on only two-states transformation - maintaining, a total austenite phase by continuous heating or maintaining a total martensite phase by continuous cooling. Due to the huge advantages provided by SMA. the Control of SMA's phase transformation phenomenon is essential for its usage in many areas of aerospace, biomedical, automobile, vibration-dampers, etc. The main hurdle is that the nonlinearity can't be modeled effectively even by 3rd-degree differential equations. Additionally, the apparatus used for measurement of strain recovery during SMA actuation is includes linear variable differential transducer (LVDT), which is too very bulky and expensive and doesn't fully utilize the potential of SMA applications in miniaturized devices. This research work presents a method to eliminate the bulky position sensor i.e. LVDT by introducing an artificial neural network (ANN). This elimination will also reduce the total cost of an SMA system. Various researchers have attempted to model the behaviour of SMAs using ANN techniques but these models still had a high RMS error. In this paper, we develop a complex ANN model to predict SMAs behaviour and estimate strain recovery based on identified input parameters.
Current is used to produce the Joule heating effect to raise the temperature which triggers the phase transformation in the SMA wire. The ANN model tries to model the hysteresis (and non-linearity) between the Resistance difference and the displacement of the SMA wire by taking resistance difference (dR) us input and predicting the displacement of the SMA wire. We have also developed another Neural Network model to predict the temperature of the SMA wire with the current running through it and the resistance as the input. The results of our research not only demonstrate the effectiveness of the Neural Networks for prediction of displacement and temperature oldie SMAs but also show how the proposed architectures have a much lesser error as compared to earlier models and arc much more effective at modeling SMAs.
Actions (login required)
 |
View Item |
Download Statistics
Download Statistics