SB
S. Banas
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Efficient Production Scheduling Under Uncertainty
Combining Constraint Programming, Heuristics, and Monte Carlo Simulations in Industrial Scheduling
Efficient production scheduling is a critical challenge in modern manufacturing, especially within complex environments characterised by strict operational constraints and inherent uncertainty. This thesis addresses the development and integration of a modular scheduling engine for Occator's OccaSee production management platform, a system previously dependent on manual or external scheduling tools. The challenge lies in balancing the computational complexity of industrial constraints—such as sequence-dependent changeovers, alternative resources, and distinct batch types—with the need for reliable decision support under stochastic disruptions.
To tackle this, a comprehensive scheduling framework was developed. First, a heuristic split-schedule-join paradigm is introduced to handle lot sizing, eliminating combinatorial explosion while preserving order-splitting flexibility. Second, a modular Constraint Programming (CP) model is formulated to optimise batch assignments and sequencing using a multi-step lexicographic approach prioritising lateness, transition costs, and total completion time. To ensure practical industrial viability and scalability, a custom greedy constructive heuristic is integrated as a warm-start mechanism for the CP solver. Finally, operational uncertainty is addressed through a Monte Carlo simulation engine that models historical duration deviations, forecasting schedule robustness and quantifying the value of reactive rescheduling.
Empirical evaluation across diverse static and disrupted instances demonstrates the framework's efficacy. The CP model assisted by the greedy warm-start (CP-WS) guarantees 100% feasibility on highly constrained instances, accelerates early convergence by a factor of 2.3, and reduces lateness and transition costs by approximately 50% compared to the greedy baseline. Furthermore, stochastic experiments reveal that while minor disruptions are best absorbed by existing schedule structures, targeted reactive rescheduling under severe delays recovers an average of over 800 minutes of accumulated lateness. By combining constraint programming, heuristic acceleration, and simulation-based risk analysis, this thesis transforms OccaSee into a dynamic decision support system capable of robust industrial scheduling. ...
To tackle this, a comprehensive scheduling framework was developed. First, a heuristic split-schedule-join paradigm is introduced to handle lot sizing, eliminating combinatorial explosion while preserving order-splitting flexibility. Second, a modular Constraint Programming (CP) model is formulated to optimise batch assignments and sequencing using a multi-step lexicographic approach prioritising lateness, transition costs, and total completion time. To ensure practical industrial viability and scalability, a custom greedy constructive heuristic is integrated as a warm-start mechanism for the CP solver. Finally, operational uncertainty is addressed through a Monte Carlo simulation engine that models historical duration deviations, forecasting schedule robustness and quantifying the value of reactive rescheduling.
Empirical evaluation across diverse static and disrupted instances demonstrates the framework's efficacy. The CP model assisted by the greedy warm-start (CP-WS) guarantees 100% feasibility on highly constrained instances, accelerates early convergence by a factor of 2.3, and reduces lateness and transition costs by approximately 50% compared to the greedy baseline. Furthermore, stochastic experiments reveal that while minor disruptions are best absorbed by existing schedule structures, targeted reactive rescheduling under severe delays recovers an average of over 800 minutes of accumulated lateness. By combining constraint programming, heuristic acceleration, and simulation-based risk analysis, this thesis transforms OccaSee into a dynamic decision support system capable of robust industrial scheduling. ...
Efficient production scheduling is a critical challenge in modern manufacturing, especially within complex environments characterised by strict operational constraints and inherent uncertainty. This thesis addresses the development and integration of a modular scheduling engine for Occator's OccaSee production management platform, a system previously dependent on manual or external scheduling tools. The challenge lies in balancing the computational complexity of industrial constraints—such as sequence-dependent changeovers, alternative resources, and distinct batch types—with the need for reliable decision support under stochastic disruptions.
To tackle this, a comprehensive scheduling framework was developed. First, a heuristic split-schedule-join paradigm is introduced to handle lot sizing, eliminating combinatorial explosion while preserving order-splitting flexibility. Second, a modular Constraint Programming (CP) model is formulated to optimise batch assignments and sequencing using a multi-step lexicographic approach prioritising lateness, transition costs, and total completion time. To ensure practical industrial viability and scalability, a custom greedy constructive heuristic is integrated as a warm-start mechanism for the CP solver. Finally, operational uncertainty is addressed through a Monte Carlo simulation engine that models historical duration deviations, forecasting schedule robustness and quantifying the value of reactive rescheduling.
Empirical evaluation across diverse static and disrupted instances demonstrates the framework's efficacy. The CP model assisted by the greedy warm-start (CP-WS) guarantees 100% feasibility on highly constrained instances, accelerates early convergence by a factor of 2.3, and reduces lateness and transition costs by approximately 50% compared to the greedy baseline. Furthermore, stochastic experiments reveal that while minor disruptions are best absorbed by existing schedule structures, targeted reactive rescheduling under severe delays recovers an average of over 800 minutes of accumulated lateness. By combining constraint programming, heuristic acceleration, and simulation-based risk analysis, this thesis transforms OccaSee into a dynamic decision support system capable of robust industrial scheduling.
To tackle this, a comprehensive scheduling framework was developed. First, a heuristic split-schedule-join paradigm is introduced to handle lot sizing, eliminating combinatorial explosion while preserving order-splitting flexibility. Second, a modular Constraint Programming (CP) model is formulated to optimise batch assignments and sequencing using a multi-step lexicographic approach prioritising lateness, transition costs, and total completion time. To ensure practical industrial viability and scalability, a custom greedy constructive heuristic is integrated as a warm-start mechanism for the CP solver. Finally, operational uncertainty is addressed through a Monte Carlo simulation engine that models historical duration deviations, forecasting schedule robustness and quantifying the value of reactive rescheduling.
Empirical evaluation across diverse static and disrupted instances demonstrates the framework's efficacy. The CP model assisted by the greedy warm-start (CP-WS) guarantees 100% feasibility on highly constrained instances, accelerates early convergence by a factor of 2.3, and reduces lateness and transition costs by approximately 50% compared to the greedy baseline. Furthermore, stochastic experiments reveal that while minor disruptions are best absorbed by existing schedule structures, targeted reactive rescheduling under severe delays recovers an average of over 800 minutes of accumulated lateness. By combining constraint programming, heuristic acceleration, and simulation-based risk analysis, this thesis transforms OccaSee into a dynamic decision support system capable of robust industrial scheduling.
Evaluating Machine Learning Approaches for Predicting Drug Response in Cancer Cells
A Comparative Analysis of Geneformer and Support Vector Machine
Accurately predicting how cancer cells respond to drug treatment is important to advance drug development. This paper presents a comparative analysis of Geneformer, a deep-learning transformer pre-trained on transcriptomic data, and Support Vector Machine. Using the Sciplex2 dataset, which includes transcriptomic data from lung cancer cells treated with three drugs, both models were trained to predict the response of cancer cells to drug treatments.
This paper investigates how Geneformer and SVM perform in predicting the treatment label of cells across different drugs and doses, which drug doses are suitable for conducting single-gene perturbation experiments, how accurately can these experiments replicate drug effects, and what are the differences in results between Geneformer and SVM regarding their ability to identify significant genes affecting drug response.
Results indicate that while SVM generally achieves higher accuracy in predicting treatment labels of cells, Geneformer demonstrates better capability in identifying genes whose perturbations mimic drug effects. Geneformer's embeddings show significant shifts towards treated cell states after single-gene perturbations, indicating a deeper understanding of gene interactions in drug response. On the other hand, SVM's predictions rely more on differential gene expression. This comparative analysis underscores the strengths and limitations of each approach in modelling complex biological systems and predicting the drug response of cancer cells. ...
This paper investigates how Geneformer and SVM perform in predicting the treatment label of cells across different drugs and doses, which drug doses are suitable for conducting single-gene perturbation experiments, how accurately can these experiments replicate drug effects, and what are the differences in results between Geneformer and SVM regarding their ability to identify significant genes affecting drug response.
Results indicate that while SVM generally achieves higher accuracy in predicting treatment labels of cells, Geneformer demonstrates better capability in identifying genes whose perturbations mimic drug effects. Geneformer's embeddings show significant shifts towards treated cell states after single-gene perturbations, indicating a deeper understanding of gene interactions in drug response. On the other hand, SVM's predictions rely more on differential gene expression. This comparative analysis underscores the strengths and limitations of each approach in modelling complex biological systems and predicting the drug response of cancer cells. ...
Accurately predicting how cancer cells respond to drug treatment is important to advance drug development. This paper presents a comparative analysis of Geneformer, a deep-learning transformer pre-trained on transcriptomic data, and Support Vector Machine. Using the Sciplex2 dataset, which includes transcriptomic data from lung cancer cells treated with three drugs, both models were trained to predict the response of cancer cells to drug treatments.
This paper investigates how Geneformer and SVM perform in predicting the treatment label of cells across different drugs and doses, which drug doses are suitable for conducting single-gene perturbation experiments, how accurately can these experiments replicate drug effects, and what are the differences in results between Geneformer and SVM regarding their ability to identify significant genes affecting drug response.
Results indicate that while SVM generally achieves higher accuracy in predicting treatment labels of cells, Geneformer demonstrates better capability in identifying genes whose perturbations mimic drug effects. Geneformer's embeddings show significant shifts towards treated cell states after single-gene perturbations, indicating a deeper understanding of gene interactions in drug response. On the other hand, SVM's predictions rely more on differential gene expression. This comparative analysis underscores the strengths and limitations of each approach in modelling complex biological systems and predicting the drug response of cancer cells.
This paper investigates how Geneformer and SVM perform in predicting the treatment label of cells across different drugs and doses, which drug doses are suitable for conducting single-gene perturbation experiments, how accurately can these experiments replicate drug effects, and what are the differences in results between Geneformer and SVM regarding their ability to identify significant genes affecting drug response.
Results indicate that while SVM generally achieves higher accuracy in predicting treatment labels of cells, Geneformer demonstrates better capability in identifying genes whose perturbations mimic drug effects. Geneformer's embeddings show significant shifts towards treated cell states after single-gene perturbations, indicating a deeper understanding of gene interactions in drug response. On the other hand, SVM's predictions rely more on differential gene expression. This comparative analysis underscores the strengths and limitations of each approach in modelling complex biological systems and predicting the drug response of cancer cells.