Mihir Hemant Kulkarni
Mihir Hemant Kulkarni has a total of 3 co-authors. Here are the top co-authors by number of shared publications:
- Vivek Khanzode
Department of Operations and Supply Chain Management, Indian Institute of Management, Mumbai, Maharashtra 400087, India
3 shared publications - Shivling M. Pise
School of Engineering, Technology & Management, Tatyasaheb Kore Institute of Engineering & Technology (Autonomous)
1 shared publication - Sourabh Devidas Kulkarni
Operations Management and Decision Sciences, FORE School of Management
1 shared publication
Latest Publications
- 8 June 2026
- Published by Taylor & Francis in Quality Management Journal
- Vol. 33 (4) , 1-19
From inspection to intelligence: A data-driven quality management system for smart manufacturing
- 8 June 2026
- Published by Taylor & Francis in Quality Management Journal
- Vol. 33 (4) , 1-19
From inspection to intelligence: A data-driven quality management system for smart manufacturing
- 9 February 2026
- Published by World Scientific Pub Co Pte Ltd in Journal of Advanced Manufacturing Systems
- p. 1-28
Decision Support Framework for Workforce Allocation in Manufacturing Using Machine Learning
Efficient workforce allocation is essential for productivity and cost control, especially in small-batch, repetitive manufacturing environments. Traditional scheduling relies on static rules and managerial intuition, causing inefficiencies, downtime, and suboptimal task assignments. A data-driven approach is needed to adapt to real-time variations in operator skills, availability, and production demand. This research develops a machine learning-based framework that predicts operator-task performance, identifies key productivity factors, and generates dynamic, interpretable task assignment recommendations, improving efficiency, reducing labor waste, and supporting informed decision-making. The methodology collects operator IDs, skills, task types, shift schedules, performance metrics, and production constraints. Data cleaning, encoding, normalization, and feature engineering capture operator-task interactions and temporal effects. Predictive modelling uses Gradient Boosting Machine, regression, and time-series analysis to forecast performance and workforce demand. K-Nearest Neighbors and rule-based logic generate real-time task assignments, while Random Forest identifies critical factors and simulates allocation scenarios. Continuous learning updates models to adapt to evolving conditions, ensuring interpretable and efficient workforce management. Findings demonstrate substantial improvements: predicted operator productivity ranges from 0.70 to 0.80, task completion averages 23-32 min, and task allocation efficiency reaches 94–95%. GBM achieves 0.87 accuracy, RMSE 0.08, [ ] 0.81, and F1-score 0.84, while the hybrid DSS improves accuracy to 0.88, reduces downtime by 48%, and increases productivity by 12%. Future scope includes reinforcement learning, real-time IoT integration, and adaptive multi-objective optimization to further enhance scheduling, predictive accuracy, and operational efficiency in complex manufacturing environments.
- 9 February 2026
- Published by World Scientific Pub Co Pte Ltd in Journal of Advanced Manufacturing Systems
- p. 1-28
Decision Support Framework for Workforce Allocation in Manufacturing Using Machine Learning
Efficient workforce allocation is essential for productivity and cost control, especially in small-batch, repetitive manufacturing environments. Traditional scheduling relies on static rules and managerial intuition, causing inefficiencies, downtime, and suboptimal task assignments. A data-driven approach is needed to adapt to real-time variations in operator skills, availability, and production demand. This research develops a machine learning-based framework that predicts operator-task performance, identifies key productivity factors, and generates dynamic, interpretable task assignment recommendations, improving efficiency, reducing labor waste, and supporting informed decision-making. The methodology collects operator IDs, skills, task types, shift schedules, performance metrics, and production constraints. Data cleaning, encoding, normalization, and feature engineering capture operator-task interactions and temporal effects. Predictive modelling uses Gradient Boosting Machine, regression, and time-series analysis to forecast performance and workforce demand. K-Nearest Neighbors and rule-based logic generate real-time task assignments, while Random Forest identifies critical factors and simulates allocation scenarios. Continuous learning updates models to adapt to evolving conditions, ensuring interpretable and efficient workforce management. Findings demonstrate substantial improvements: predicted operator productivity ranges from 0.70 to 0.80, task completion averages 23-32 min, and task allocation efficiency reaches 94–95%. GBM achieves 0.87 accuracy, RMSE 0.08, [ ] 0.81, and F1-score 0.84, while the hybrid DSS improves accuracy to 0.88, reduces downtime by 48%, and increases productivity by 12%. Future scope includes reinforcement learning, real-time IoT integration, and adaptive multi-objective optimization to further enhance scheduling, predictive accuracy, and operational efficiency in complex manufacturing environments.
- 5 April 2025
- Published by World Scientific Pub Co Pte Ltd in Surface Review and Letters
COMPARATIVE ANALYSIS OF STRENGTH PROPERTIES IN 3D-PRINTED PLA AND ABS STRUCTURES USING DNN-WOA PREDICTION MODEL
In the realm of additive manufacturing, the quality and mechanical performance of 3D-printed structures are heavily influenced by the parametric settings used during the printing process. Some of these conditions include layer thickness, printing speed, infill density, filling pattern, printing substance, and so on. This research investigates the effect of printing factors on the strength characteristics of 3D-printed acrylonitrile butadiene styrene (ABS) as well as polylactic acid (PLA) structures. The parameters considered for the proposed study include material type, infill direction, infill density, and infill pattern. These four conditions are optimized using the machine-learning algorithm to obtain maximum strength in the resulting structures. The breaking load, extension, tensile strain, and tensile strength are the strength factors determined by the experiments. Furthermore, a deep neural network integrated with a walrus optimization algorithm (DNN-WOA)-based hybrid machine learning method is used in the experimental data to discover the optimal parametric conditions for constructing the structures with maximum strength. Based on the findings, infill density is a significant component in increasing both the tensile strength and elastic modulus of printed samples. Line and triangular-type infill patterns exerted the range of tensile strength with 10–20 MPa of deviations. In terms of tensile strain, the line pattern produced significantly more tensile strain than the triangle pattern. The highest breaking stress is observed at a [ ] raster angle, regardless of infill density. When compared to the strengths of ABS and PLA, the PLA with a triangle infill pattern outperformed the ABS structure printed with a line-type infill pattern. The optimum printing settings for a PLA structure with a line-type infill pattern are [ ] and 25% for raster angle and infill density, respectively. This parameter resulted in enhanced strength performance, with 45.81[ ]MPa of tensile strength, 12.13% of tensile strain, 359.34[ ]kgf of breaking load, 7.21[ ]mm of extension, and 24.12[ ]min of printing time.
- 5 April 2025
- Published by World Scientific Pub Co Pte Ltd in Surface Review and Letters
COMPARATIVE ANALYSIS OF STRENGTH PROPERTIES IN 3D-PRINTED PLA AND ABS STRUCTURES USING DNN-WOA PREDICTION MODEL
In the realm of additive manufacturing, the quality and mechanical performance of 3D-printed structures are heavily influenced by the parametric settings used during the printing process. Some of these conditions include layer thickness, printing speed, infill density, filling pattern, printing substance, and so on. This research investigates the effect of printing factors on the strength characteristics of 3D-printed acrylonitrile butadiene styrene (ABS) as well as polylactic acid (PLA) structures. The parameters considered for the proposed study include material type, infill direction, infill density, and infill pattern. These four conditions are optimized using the machine-learning algorithm to obtain maximum strength in the resulting structures. The breaking load, extension, tensile strain, and tensile strength are the strength factors determined by the experiments. Furthermore, a deep neural network integrated with a walrus optimization algorithm (DNN-WOA)-based hybrid machine learning method is used in the experimental data to discover the optimal parametric conditions for constructing the structures with maximum strength. Based on the findings, infill density is a significant component in increasing both the tensile strength and elastic modulus of printed samples. Line and triangular-type infill patterns exerted the range of tensile strength with 10–20 MPa of deviations. In terms of tensile strain, the line pattern produced significantly more tensile strain than the triangle pattern. The highest breaking stress is observed at a [ ] raster angle, regardless of infill density. When compared to the strengths of ABS and PLA, the PLA with a triangle infill pattern outperformed the ABS structure printed with a line-type infill pattern. The optimum printing settings for a PLA structure with a line-type infill pattern are [ ] and 25% for raster angle and infill density, respectively. This parameter resulted in enhanced strength performance, with 45.81[ ]MPa of tensile strength, 12.13% of tensile strain, 359.34[ ]kgf of breaking load, 7.21[ ]mm of extension, and 24.12[ ]min of printing time.