Forecasting Material Quantity Using Machine
Learning and Time Series Techniques
MUHAMMAD RISHAD ES5
CONTENTS
1. Introduction
2. Motivation and Contribution
3. Literature Review
4. Methods
5. Materials
6. Results and Discussion
7. Conclusion
01
INTRODUCTION
INTRODUCTUON
● In every production system, human operators play a crucial role in operations, adjustments, and maintenance.
● However, the advancement of technology has paved the way for automated production systems, aiming to minimize
human intervention and optimize resource utilization.
● Rapid technological changes have presented numerous challenges for factories; however, they have also brought about
various emerging technologies such as the Internet of Things (IoT), big data, cloud computing, artificial intelligence,
and its associated techniques.
● The concept of Industry 4.0 has emerged as a new paradigm for industrialization, leveraging these technologies to
address these challenges effectively.
● At the heart of Industry 5.0 lies the concept of the smart factory which is used for process optimization, production
monitoring, control, and predictive maintenance.
● Smart factories utilize these cutting-edge technologies to enhance the performance, quality, control, and transparency of
the manufacturing process.
02
Motivation and Contribution
Motivation and Contribution
Details about the data, the
This study makes a valuable
industrial process, and the The developed model proved
contribution to the
deep learning techniques its efficiency in terms of
industrial field such as
applied in the study are constructing intelligent
Cement Production, because
highlighted, in addition to models that allow optimized
of the several advantages
the focus on findings and participation more than
that the work highlights.
their economic impact on human operators.
the real system.
03
Literature Review
Literature Review
● Forecasting is a critical aspect of supply chain management and production planning, as accurate predictions of
material quantities can help organizations optimize inventory, reduce costs, and enhance overall efficiency.
● Machine Learning (ML) methods have been proposed in the academic literature as alternatives to statistical ones for
time series forecasting.
● One of the most powerful machine learning techniques is Random Forest (RF).
● The present literature review explores many applications of deep learning techniques, including LSTM and GRU in
forecasting since their introduction.
04
Methods
Methods
● The paper discusses the intricacies of data collection, addressing challenges related to data quality and availability.
● This research employs a suite of machine learning algorithms for the prediction of the quantity of material needed for
the production of clinker, including Support Vector Machine (SVM), Random Forest (RF), Multi-Layer Perceptron
(MLP), and Gradient Boosting.
● Moreover, time series techniques including LSTM and GRU are also selected for this task.
● All numerical features were normalized before training to ensure consistent input scale and improve model
performance.
05
Materials
Materials
● The study methodology, feature engineering approaches, model development, and extensive discussions on the
implications of our findings are based on the rich dataset that allows the application of various machine-learning
models to predict the amount of the materials.
● The selected workshop for investigation is the cement mill workshop in the Ain Touta cement factory (SCIMAT) located
in the East of Algeria.
● The cement mill workshop plays a crucial role in the cement production line.
Parameters Interval Units Designation
P1L01_L21 [0-100] % Silo Cement Level
P1L02_L21 [0-100] % Silo Cement Level
P2L01_L21 [0-140] t/h [Link] Flow
Z2M01_P2 [0-40] mbar Crusher Pressure Output
Z2M01_T2 [0-150] °C Crusher Temperature Input
06
Results and Discussion
Results and Discussion
The data set is split into two parts,
respectively the train set (60%), test set
(30%), and validation set (10%).
The results of the evaluation process using
the Machine Learning algorithm provide
valuable insights into the predictive
capabilities of the machine learning
models and their ability to accurately
forecast the state of the production line.
07
Conclusion
Conclusion
1 The presented learning models and architectures 2 Future research should focus on testing the presented
showcase significant advancements that bring notable dataset with a wider range of advanced deep-learning
improvements in control flexibility, data handling algorithms to identify the most suitable approach for
capabilities, and the ability to process large amounts of specific industrial processes.
information in complex industrial processes.
Thank You