Predicting Of Proactive Environmental Management For Unhairing Wastewater Treatment In Tunisia Using Neural Network Learning Algorithms

Purpose The purpose of this study is to focus on Tunisian tannery sector that causes a considerable damage to the environment and consequently leads to serious health problems due to the untreated effluents generated from the various leather processing stages. Design/methodology/approach This paper discusses a voluntary initiative taken by the top managers of tannery enterprise to prevent pollution and disseminate the concept of eco-industrial activities between employees and stakeholders. In addition, this research assesses the performance of such treatment that characterizes the chemical parameters of generated pollutants. It also aims at optimizing the industrial process for cleaner production. Coagulation-flocculation process is investigated in this study. Moreover, oxidation phase by ozone is taking into account before and after coagulation-flocculation process to measure the effectiveness of the combined method for reducing the main pollutant concentrations. Findings The unhairing and chrome (Cr) tanning steps are considered the most polluting steps. Therefore, the application of various treatment techniques, including chemical and physicochemical processes, is realized to reduce the toxicity of the effluents. The correlation between experimental and modeling results, using artificial neural network (ANN) method, was investigated in this research. The results of the constructed ANN model are measured by the correlation of experimental and model results during coagulation-flocculation and oxidation stages. The validation of the elaborated model through the error calculation (MSE) and the correlation coefficient (R) confirm the reliability of ANN method. Originality/value Eventually, the establishment of ANN model for performance prediction of wastewater parameters is investigated due to different measurements of physical effluent outputs, such as: pH, turbidity, TSS, DS, COD, fat, TSS, S2- and Cr. This study uses predictive modeling, a machine learning technique to tackle the problem of accurately predicting the behavior of unseen configuration.