The paper outlines a do-it-yourself optical sensor system for measuring water colour that can be used in citizen science projects to engage laypersons in science. The automated optical sensor system uses a camera with a Raspberry Pi computer to determine water colour according to the Forel–Ule colour scale. The collected data can be shared to the EyeOnWater database and used to provide insight and knowledge for further research in the fields of limnology and oceanology.

This project developed a low-cost, real-time portable device for measuring radiation. The device is based on a camera similar to those used in modern smartphones and includes the ability to connect to the internet, allowing users to access data through any web browser. The program files are available for download for educational purposes.

Globally, water has been the cause of the rise of civilization and also a cause of conflict. In this changing climate, managing water, the most precious commodity, has become a necessity. If we want to manage a resource properly, we need technology and data. With the innovative WAMO 300, a decentralised solution for this problem has been provided. WAMO 300, short for Water Monitor 300, acts as an IoT (Internet of Things) system that brings data from the physical world to the digital world.

Robust and interpretable machine learning algorithms, which have been proven in various applications, often lack efficient implementation on limited hardware. The novel approach is to convert the inference of the interpretable ML into a deep neural network, which is efficiently executable on edge hardware. This approach was validated in terms of runtime efficiency, memory requirements, and accuracy and resulted in a significant improvement in terms of runtime and memory requirements.

In a miniaturized gas chromatography system, there could be identification doubt between compounds eluting at close retention times. When a sensor is a thermal conductivity detector (TCD), we investigated a double-measurement approach (two temperatures) to improve specificity. Unfortunately, both theory and experiments showed that this method could only work for relatively high compound concentrations (1–50 ppm). This is therefore not relevant for a trace analysis system (1–10 ppb).

In this study, we consider the planform of a cantilever piezoelectric vibrating energy harvester (PVEH) device in which the inertia of the cantilever itself is dominant relative to the inertia of an edge block. We derive the optimal planform which ensures the harvested energy is maximal. We show that the contours of the optimal planform are defined by Bessel functions. The predictive capabilities of our model are demonstrated by comparison to finite-element simulations.

In many investigations of piezoelectric energy harvesters, it is often tacitly assumed that optimal performance is achieved when the vibration mode of the structure ensures a uniform strain within the piezoelectric layer. This manuscript provides, for the first time ever, a rigorous proof of this point.

We present a sensor with an original architecture for gas sensing that is 4 times more efficient than other sensors based on the same detection technology. It has a small volume (less than 1 cm³), low cost, and low power consumption, making it possible for it to be used in compact systems and even in smartphones. It is able to detect small changes in carbon dioxyde concentrations in ambient air for indoor air quality monitoring.

We present a new method for creating flexible temperature sensors that can be applied to complex shapes. By using a technique called water transfer printing, these sensors can be transferred directly onto surfaces without the need for flexible films . This allows them to be more accurate, more reliable, and more adaptable than traditional sensors.  These sensors can work well in various environments and are unaffected by humidity, making them ideal for use in healthcare, robotics, and wearable devices.

Selecting the best model for monitoring industrial machines is challenging due to limited data and varying conditions. While deep learning (DL) is often considered ideal, it does not always perform well. We compared DL with traditional methods focused on feature extraction and classification. Our tests on seven datasets revealed that traditional methods are more reliable when machine conditions change, highlighting the importance of simpler, interpretable models for industrial monitoring.