ABSTRACT

In this paper, coal screening in vibrating screen was carried out with the size ranges of −6 mm + 4 mm, −4 mm + 2 mm, and −2 mm + 0.5 mm. The vibrating screen was newly designed with flexibility in angle and frequency. The vibrating screen experimentation was carried out by varying screen mesh, angle, and screen frequency. During the screening, the angle was kept constant, and frequency was varied to obtain each size range’s screening efficiency. The experimental results of screening efficiency were evaluated for each size fraction range of coal. The maximum efficiency for screening coal with −6 mm+4 mm, −4 mm+2 mm, and −2 mm+0.5 mm size range obtained was 87.60%, 80.93%, and 62.96%, respectively. Further, the prediction model was developed for each size range using a feed-backward artificial neural network (ANN) to consider the back-propagation error technique. For each screening condition, 10 ANN models were developed with the variation in 1–10 different neurons. ANN has provided mathematical models with a 99.9% regression coefficient for predicting each size range’s screening efficiency. Furthermore, the residuals of each optimal ANN model were analyzed using a normal probability plot and histogram. The ANN model’s accuracy was obtained from the residual analysis by evaluating four different model conditions, i.e., independence, homoscedasticity, normality, and mean error.

摘要

本文对振动筛中的煤进行筛分，粒度范围为-6 mm + 4 mm、-4 mm + 2 mm 和-2 mm + 0.5 mm。振动筛经过全新设计，角度和频率灵活。振动筛实验是通过改变筛孔、角度和筛频进行的。在筛分过程中，角度保持不变，通过改变频率来获得每个尺寸范围的筛分效率。对煤的每个粒度分数范围的筛选效率的实验结果进行了评估。获得的-6 mm+4 mm、-4 mm+2 mm和-2 mm+0.5 mm粒度范围煤的最大筛分效率分别为87.60%、80.93%和62.96%。更远，使用反馈人工神经网络 (ANN) 为每个尺寸范围开发预测模型，以考虑反向传播误差技术。对于每个筛选条件，开发了 10 个 ANN 模型，其中包含 1-10 个不同神经元的变化。ANN 提供了具有 99.9% 回归系数的数学模型，用于预测每个尺寸范围的筛选效率。此外，使用正态概率图和直方图分析每个最佳 ANN 模型的残差。通过评估四种不同的模型条件，即独立性、同方差性、正态性和平均误差，从残差分析中获得了人工神经网络模型的准确性。9% 回归系数，用于预测每个尺寸范围的筛选效率。此外，使用正态概率图和直方图分析每个最佳 ANN 模型的残差。通过评估四种不同的模型条件，即独立性、同方差性、正态性和平均误差，从残差分析中获得了人工神经网络模型的准确性。9% 回归系数，用于预测每个尺寸范围的筛选效率。此外，使用正态概率图和直方图分析每个最佳 ANN 模型的残差。通过评估四种不同的模型条件，即独立性、同方差性、正态性和平均误差，从残差分析中获得了人工神经网络模型的准确性。