The present research presents a framework that supports the development and operation of machine-learning (ML) algorithms to develop, maintain and manage the whole lifecycle of modeling software sensors related to complex chemical processes. Our motivation is to take advantage of ML and edge computing and offer innovative solutions to the chemical industry for difficult-to-measure laboratory variables. The purpose of software sensor models is to continuously forecast the quality of products to achieve effective quality control, maintain the stable production condition of plants, and support efficient, environmentally friendly, and harmless laboratory work. As a result of the literature review, quite a few ML models have been developed in recent years that support the quality assurance of different types of materials. However, the problems of continuous operation, maintenance and version control of these models have not yet been solved. The method uses ML algorithms and takes advantage of cloud services in an enterprise environment. Industrial 4.0 devices such as the Internet of Things (IoT), edge computing, cloud computing, ML, and artificial intelligence (AI) are core techniques. The article outlines an information system structure and the related methodology based on data from a quality-assurance laboratory. During the development, we encountered several challenges resulting from the continuous development of ML models and the tuning of their parameters. The article discusses the development, version control, validation, lifecycle, and maintenance of ML models and a case study. The developed framework can continuously monitor the performance of the models and increase the amount of data that make up the models. As a result, the most accurate, data-driven and up-to-date models are always available to quality-assurance engineers with this solution.

In the context of hierarchical system modeling, ensuring constraints between different hierarchy levels are met, so, for instance, ensuring the aggregation constraints are satisfied, is essential. However, modelling and forecasting each element of the hierarchy independently introduce errors. To mitigate this balance error, it is recommended to employ an optimal data reconciliation technique with an emphasis on measurement and modeling errors. In this study, three different machine learning methods for development were investigated. The first method involves no data reconciliation, relying solely on machine learning models built independently at each hierarchical level. The second approach incorporates measurement errors by adjusting the measured data to satisfy each constraint, and the machine learning model is developed based on this dataset. The third method is based on directly fine-tuning the machine learning predictions based on the prediction errors of each model. The three methods were compared using three case studies with different complexities, namely mineral composition estimation with 9 elements, forecasting of retail sales with 14 elements, and waste deposition forecasting with more than 3000 elements. From the results of this study, the conclusion can be drawn that the third method performs the best, and reliable machine learning models can be developed.