Simultaneous determination of Copper, Zinc and Nikel in Electroplating Waste water by UV-VIS Spectroscopy Combined with advanced Machine Learning and Deep Learning Models

Abstract

Monitoring to evaluate wastewater quality during the production process requires simple measurements and Realtime analysis as well. Among common methods for heavy metal analysis, the UV-VIS absorption spectroscopy is considered a potential analytical method due to its low cost and simple operation, direct online integration with treatment tanks. However, it faces limitations in simultaneously analyzing multiple metals due to overlapping absorption spectra. This study applied machine learning (ML) algorithms (Decision Tree (DT), Random Forest(RF) and deep learning (DL) models (Multilayer Perceptron - MLP, and 1D Convolutional Neural Network - 1D-CNN) to improve the accuracy of simultaneous quantitative analysis of three metals—Cu, Zn, and Ni in electroplating wastewater—based on VIS absorption spectra data of their colored complexes in aqueous solution with the PAN reagent in the presence of a surfactant. Large datasets were collected from UV-VIS spectra of 500 wastewater spiked samples in the range of 620-500 nm with a 1 nm interval, resulting in a dataset of size 500x121, followed by the application of ML and DL models using the Python programming language. Model performance was evaluated based on the correlation coefficient (R²) and root mean square error (RMSE). Preprocessing methods such as first-order derivatives and Principal Component Analysis (PCA) were applied to reduce noise in the dataset before training with machine learning algorithms. Results showed that the 1D-CNN model outperformed the others, achieving R² > 0.88 and RMSE < 0.036 for all three analytes. It is supposed by its ability to directly extract nonlinear features from raw data without the need for dimensionality reduction. In contrast, the DT, RF, and even MLP models, which utilized principal component analysis (PCA) for dimensionality reduction, demonstrated significantly lower accuracy due to information loss during the reduction process. The proposed model was successfully applied for rapid and simple metal concentration determination in practical samples using a test kit with reagent, a compact spectrophotometer, and an automated PC-based data reading application. These findings demonstrate that combining UV-VIS spectroscopy with machine learning and deep learning algorithms is an effective and feasible approach for the simultaneous detection of multiple heavy metals in specific matrix wastewater samples.