Micro dosing has a wide range of applications, specifically in medical treatment. The need for a slow or constant rate of drug delivery in a safe range of accuracy and precision is among the main requirements in drug delivery systems [Bußmann et al., 2021a]. It was shown that using a continuous drug delivery system in insulin therapy for patients who have diabetes can improve their health compared to regular treatments [Jeitler et al., 2008]. Patch pumps can be used as portable devices and, as a result, provide continuous drug delivery for better treatment [Bußmann et al., 2021b]. Compared to standard solutions, the cost efficiency of patch pumps is vital because at least some parts are disposable.
As part of a patch pump, a micropump, specifically a piezoelectric pump, can save more space and provide energy-efficient and lightweight systems for portable drug delivery applications [Zhang et al., 2013]. Different pump conditions, such as changes in backpressure or reservoir over/under pressure and the transfer of air bubbles, can emerge during a micro-dosing application. As a result, to attain higher flow rate accuracy in piezoelectric actuation, not only applied voltage and frequency are crucial, but any change in pump condition must be considered. The first approach is to deploy more sensors to reach an acceptable flow rate accuracy. Any additional sensors, such as pressure sensors, increase the cost and complexity of the whole system. Piezoelectric materials simultaneously have sensory (direct) and actuator (indirect) effects.
Aim of the project:
In this thesis, a new approach is considered to use direct and indirect piezo effects simultaneously. Using both effects provides an efficient way to investigate any change in pump condition without using any additional sensors. Piezo sensory (self-sensing) signal is superimposed with the charging current of the piezoelectric. It contains valuable information in different conditions, such as backpressure and air bubble transmission. Each condition creates specific characteristics in different parts of the signal. However, using just general characteristics of self-sensing signals leads us to inaccurate condition detection methods. Nevertheless, machine learning methods in a higher dimensional feature space provide us with much more accurate condition detection methods.
The main goal of this work is to estimate backpressure in a piezoelectric micropump. Machine learning methods will be implemented and evaluated on limited hardware to estimate different backpressure levels using the self-sensing signal. To achieve this goal, the following steps will be performed:
1. Obtain the training data by measuring the self-sensing signals and related changes in pump conditions.
2. Preprocess and label the samples in the dataset.
3. Train machine learning and deep learning methods to estimate different backpressure levels.
4. Evaluate the performance of the trained models on the target device.
- [Bußmann et al., 2021a] A. B. Bußmann, L. M. Grünerbel, C. P. Durasiewicz, T. A. Thalhofer, A. Wille, and M. Richter. Microdosing for drug delivery application—a review. Sensors and Actuators A: Physical, 330:112820, 2021. doi: 10.1016/j.sna.2021.112820.
- [Bußmann et al., 2021b] Bußmann, Agnes, et al. “Piezoelectric silicon micropump for drug delivery applications.” Applied Sciences 11.17 (2021): 8008.
- [Jeitler et al., 2008] K. Jeitler, K. Horvath, A. Berghold, T. W. Gratzer, K. Neeser, T. R. Pieber, and A. Siebenhofer. Continuous subcutaneous insulin infusion versus multiple daily insulin injections in patients with diabetes mellitus: systematic review and meta-analysis. Diabetologia, 51(6):941–951, 2008. doi: 10.1007/s00125-008-0974-3.
- [Zhang et al., 2013] Z. Zhang, J. Kan, G. Cheng, H. Wang, and Y. Jiang. A piezoelectric micropump with an integrated sensor based on space-division multiplexing. Sensors and Actuators A: Physical, 203:29–36, 2013. doi: 10.1016/j.sna. 2013.08.027.