Experimental analysis and data-driven machine learning modelling of the minimum ignition temperature (MIT) of aluminium dust. (15th September 2022)

• Record Type: Journal Article
• Authors: Arshad, Ushtar; Taqvi, Syed Ali Ammar; Buang, Azizul
• Is Part Of: Fuel. Volume 324:Part A(2022)
• Publication Date: 2022-09-15
• Subjects: Dust explosion -- MIT -- Godbert Greenwald furnace -- Aluminium dust -- ANN; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2022.124569
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: Experimental investigation of the MIT of aluminium dust using Godbert-Greenwald furnace. Development of predictive model for MIT using ANNs. Correlation developed using polynomial surface fitting to predict the MIT. Effects of dust cloud's concentration and dispersion pressure on MIT. ANN and PSF approaches for prediction and comparison with the unseen test data set of MIT. Abstract: The industrial sector continues to face difficulties in preventing dust explosions. Ignition, in particular, is a phenomenon that has yet to be fully comprehended. As a result, safety conditions pertaining to ignition control are rarely assessed to an adequate level. It is generally recognised that the ignition behaviour of combustible dust is influenced by a variety of parameters, including the chemical composition, particle size, moisture content, dispersion pressure, the concentration of dust and so on, but there is still a lack of understanding regarding the simultaneous effect of multiple influential variables. This article aims to provide data on the minimum ignition temperatures of combustible dust using aluminium dust. The minimum ignition temperatures (MITs) were evaluated in a Godbert-Greenwald (GG) furnace with synergistic effects of dispersion pressure and concentrations for two distinct particle size ranges. Based on the statistical nature of the dust explosions and controlling parameters, this study uses data-driven modelling approaches. The experimental data has been divided into the training set and testing set in the proportion of 85% (for training) and 15% (for testing) respectively. A machine learning artificial neural network approach with Levenberg-Marquardt algorithm is implemented to obtain the predictive model for MIT of aluminium dust for both the particle size ranges (100–63 µm, 50–32 µm). The resultant model was obtained with acceptable accuracy in terms of both the training and test data sets. Besides, a statistical surface fit approach has also been adopted to model the MIT to obtain the correlations. It was found that the predictive accuracy is significantly higher for the developed ANN model than the surface fitting based on the minimum values of AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). The assessment of the MIT for combustible dust is crucial in preventing the ignition and subsequent dust explosion. If a sufficiently accurate estimate of MIT is available then the temperature of the surrounding equipment in the process industries can be controlled well below that particular value.