Artificial Intelligence for pollutant recognition applied to smart sensors based on SENSIPLUS

The problem of detecting pollutants in wastewater is of fundamental importance for public health and security. In a fast-developing paradigm, such as the one represented by the Smart City, mapping wastewater systems by detecting pollution sources is a required task that if properly treated helps protect ecosystems and may allow for recovering energy, recoverable material, and nutrients. Generally, this is made by involving the usage of laboratory-based analyses performed by expert professionals that ends to result in high costs and time spending approach which does not allow for timely prevention of environmental disasters. In this regard, to allow an effective prevention activity, a large number of distributed measurement systems are required. In this Ph.D. thesis are presented two possible solutions based on Machine Learning (ML) techniques using the same sensing part. The proposed measurement systems are based on the so-called Smart Cable Water (SCW) sensor, a multi-sensor based on SENSIPLUS technology developed by Sensichips s.r.l. More in detail, one solution is aimed at the development of an end-to-end IoT-ready system for the recognition of a set of substances able to reduce the false positive samples by distinguishing the outlier from the interest ones. In this regard, the system is composed of three functional blocks: a Finite State Machine (FSM) to correctly detect the substance’s passage, an anomaly detection classifier to reject all the outlier samples, and a multiclass classifier to correctly recognize the given substance. It is important to note that the capability to distinguish between outlier (not of interest) and inlier (of interest) substances drastically improves the classification performance, especially in terms of false positive rates. An extensive experimental campaign on different contaminants has been carried out to train machine learning algorithms suitable for low-cost and low-power Micro Controlled Unit (MCU). The obtained results demonstrate an excellent classification ability, achieving an accuracy of more than 95% on average, and are a reliable “proof of concept” of a pervasive IoT system for distributed monitoring. The other solution is aimed at the development of an edge computing classifier to be implemented abord the SCW sensor. In this regard has been used the Principal Component Analysis (PCA) decomposition to project the acquired data from a 10-dimensional space to a 3-dimensional one. Next, has been developed an ad-hoc classifier capable to distinguish contaminants in the projected 3-dimensional space. To learn the best classifier’s parameters has been used an evolutionary algorithm. The proposed system achieved the best accuracy of 83%, outperforming the other state-of-art systems compared. The novelty of the proposed system lies in the usage of an evolutionary algorithm for the optimization of the parameters of a novel PCA-based classification algorithm capable to detect and recognize a set of wastewater pollutants.