In the bioengineering and biochemical research areas, the Polymerase Chain Reaction (PCR) is the prerequisite step for gene chips preparation. In the PCR reaction system, the temperature in the process of PCR thermal cycling must be very strictly controlled. Through analyzing the characteristics of the heating and cooling subsystems, the researchers in this project experimentally completed the control algorithms based on the conventional PID and predicative control strategies. In this PCR temperature control system, for the given temperature-time pattern curve, to conventional PID control algorithm, the temperature maximum overshoot is less than 3degC, the stable accuracy of temperature control is plusmn1degC, the maximum heating and cooling speeds are both 2degC/s. However, the experimental results demonstrated that the PCR thermal cycler based on predictive control algorithm has better temperature tracking capability, the maximum heating and cooling speeds are twice quick and the stable error is only one half as those of ordinary PID based PCR thermal cyclers and can meet the demands of control qualities for the PCR gene-chips preparation in some situations.
With the proliferation of Smart Home and Artificial Intelligence of Things (AIoT), humanized while effective HumanComputer Interface (HCI) has been the central focus of research to bridge user’s daily life and ever-growing hardware capability. In this context, ultrasound-based acoustic sensing is being widely adopted due to its nature of contact-less and platform-agnostic. In order to fully exploit the potential of ultrasonic bandwidth and kill two birds with one stone, we propose SignID, a multi-task learning boosted system capable of jointly recognizing user gestures and identification. The system only needs a pair of speaker and microphone, which is available on various smart devices, to emit sound wave and receive the Doppler-shifted echo. To handle ambient noise and lower the computation overheads, we adopt a bandpass filter to remove the frequency band out of interest, then apply undersampling to compress the measurement for the subsequent FFT processing. After segmentation and spectrogram extraction, a multi-task deep neural network based on cross-stitch units is responsible for jointly learning representation for both identification and gesture recognition. The proposed system is evaluated based on our implementation in a real-life scenario, and the overall accuracy is over 95%. We also test the impact of different environmental factors, including user distance, ambient noise, body gesture, and training data volume. The results prove our model’s robustness to various disturbances.
The remote periodic message not only has timing constraints but also often has precedence constrains in Foundation Fieldbus system. In this paper, analysis and heuristic scheduling for remote periodic messages are proposed. Firstly, precedence constrains are considered and described by modifying release times and deadlines of remote periodic messages. Secondly, a heuristic algorithm for building FF schedule time list is proposed to find acceptable scheduling solutions by some simple strategies in order to meet these messages' timing constraints. Finally, an example of application shows the process of building FF schedule time list.