Abstract

Machine learning can be effectively used to generate models capable of representing the dynamic of production processes of small and medium-sized enterprises. These models enable the estimation of key performance indicators, and are often used for optimizing production processes. However, in most industrial applications, modeling and optimization of production processes are currently carried out as separate tasks, manually in a very costly and inefficient way. Automated machine learning tools and frameworks facilitate the path for deriving models, reducing modeling time and cost. However, optimization by exploiting production models is still in infancy. This work presents a methodology for integrating a fully automated procedure that embraces automated machine learning pipelines and a multi-objective optimization algorithm for improving the production processes, with special focus on small and medium-sized enterprises. This procedure is supported on embedding the generated models as objective functions of a reference point based non-dominated sorting genetic algorithm, resulting in preference-based Pareto-optimal parametrizations of the corresponding production processes. The methodology was implemented and validated using data from a manufacturing production process of a small manufacturing enterprise, generating highly accurate machine learning-based models for the analyzed indicators. Additionally, by applying the optimization step of the proposed methodology it was possible to increase the productivity of the manufacturing process by 3.19 % and reduce its defect rate by 2.15 %, outperforming the results obtained with traditional trial and error method focused on productivity alone.

Introduction

Nowadays, generalized adoption of the new manufacturing paradigms involves the assimilation of key technologies such as intelligent data analysis and machine learning (ML), among others, aiming at the digital transformation of enterprises [ 46 , 56 ]. This digital transformation is crucial to keep up with the competition, especially in the case of small and medium-sized manufacturing enterprises (SMEs). One contemporary target is to create a highly reconfigurable, decentralized, dynamic, self-organizing, and real-time (or near real-time) decision-making infrastructure enabling to analyze customer expectations and reach their targets [ 20 ]. By applying these transformations, SMEs should be capable of monitoring and improving their key performance indicators (KPIs) [ 67 ]. However, in practice, SMEs face several difficulties in applying these technologies, mostly related to the transition and maintenance costs, innovation complexity, and personnel training [ 50 ]. Additionally, due to limited human and computational resources, time constraints and complexity of the optimization processes, SMEs usually focus their efforts on a single productivity objective, despite being more desirable to consider and optimize multiple objectives, which leads not only to more efficient but also a more sustainable and environmentally friendly production. Due to these obstacles, a large number of SMEs don’t count yet with the necessary tools to continue with their digital transformation. In this context, the development of tools for generating useful information and smart recommendations of production systems in SMEs is almost mandatory in a high-competitive market and has a large number of potential adopters [ 12 , 42 ].

Conclusions

This work presents an automated machine learning methodology for optimizing manufacturing processes in SMEs by combining the standard tasks of AutoML tools, such as data preprocessing, feature selection, model training, and hyperparameter optimization, with preference-based multi-objective optimization. For this purpose, the basic AutoML workflow is used to generate models for each of the KPIs of the production process and, then, a new automated optimization step is introduced for using the generated models as objective functions, resulting in optimal parametrizations of the production process. By simplifying the way of interacting with this methodology, it is possible that manufacturing SMEs with low availability of highly-skilled personnel or limited computing power can benefit from advanced technologies making easier the digitalization and application of Industry 4.0 paradigm.

The methodology was implemented and validated in a production process where, firstly, the most relevant features for modeling each key performance indicator were automatically selected based on the Pearson’s correlation coefficient, allowing to reduce the dimensionality of the data. Then, models of key performance indicators were generated and their architecture/hyperparameters optimized. Generated models were compared to models obtained through other AutoML frameworks offering similar results, with values of MSE = 2.585 and R2 = 0.999, and MSE = 6 × ۱۰−۵ and R2 = 0.735, respectively. Finally, the models were used as objective functions in the R-NSGA-II algorithm for finding optimal parametrizations of the production process, yielding an improvement in both KPI, reducing scrap by 2.15 % and increasing throughput by 3.19 %, with regard to the baseline of conventional parametrization considering only a single productivity target. These improvements contribute to a higher production rate while, at the same time, the number of defective components is reduced, which underscore the potential of the proposed methodology to significantly boost overall efficiency and profitability for SMEs by optimizing their production processes more holistically.