The dissertation is devoted to the development and research of methods and models for assessing, predicting and minimising metrological risks that affect the quality of products at the manufacturing stage.
The thesis consists of an introduction, four chapters, conclusions, a list of references and appendices.
The introduction provides a justification for the relevance of the research topic, in particular the importance of managing metrological risks to ensure product quality at the manufacturing stage. The object and subject of the study are defined, the aim and objectives of the work are formulated, the scientific novelty of the results obtained and their practical significance are outlined. The article also provides an overview of the current state of the problem and the key areas that were studied in the framework of the work.
The first section presents a thorough analysis of the current state of the theory and practice of managing metrological risks and their impact on product quality at the manufacturing stage. The article defines what a metrological risk is, as well as the sources of its occurrence in the context of production processes. Particular attention is paid to the risk management process, which includes identification, assessment and monitoring of possible hazards. The article also analyses the main international standards governing risk management, in particular ISO 9000, ISO 31000, AS4360, COSO ERM, CoCo and IRM, which establish the principles and methods of effective risk management in various industries. The process of risk management in accordance with the above standards is described in detail, in particular at the stages of planning, implementation, verification and adjustment of risk mitigation measures.
In the second section, the factors of influence on the technological process as a source of metrological risks are studied, and their classification into internal and external is carried out. The specifics of the system for assessing metrological risks to product quality at the manufacturing stage, which meets the requirements of international standards (ISO 9001, ISO 31000, ISO 31010), are investigated. The concept of metrological risk management is developed, consisting of three key stages: planning, assessment and processing of risks. At the planning stage, the goals, scope, process participants and risk criteria are determined. During the assessment stage, risks are analysed, identified and evaluated. The paper presents a generalised flowchart of the process of identification and analysis of metrological risks. Additionally, a graph model of complex metrological risk is used to visualise the levels of interaction and interrelationships of various factors and indicators that form the overall level of risk. At the final stage, risks are minimised by analysing the proposed measures and their impact on the identified risks.
In the third section, a mathematical model is developed to determine a comprehensive indicator of the level of metrological risk. The scales for assessing the significance of consequences, probability of occurrence and identification of risks are proposed, which serve as the basis for determining their value. An influence matrix has been developed to determine the weighting factors that take into account the interaction of risks when determining a group risk indicator. The article also presents an adaptive model that combines group indicators into a single complex indicator and approaches to establishing the permissible value of this indicator, which allows quantifying the level of metrological risk and determining the critical limits for the production process.
It is proposed to evaluate the effectiveness of measures to minimise risks, focusing on two key criteria: the relative reduction of the complex risk indicator and the ratio of the cost of implementing measures to the possible losses that were avoided. This makes it possible to reasonably choose the optimal strategies for managing metrological risks and increase the efficiency of production processes.
Section 4 presents a method for predicting metrological risks using neural networks for time series analysis. Six models were investigated, in particular: Facebook Prophet, Statsmodels SARIMAX, Forecaster Recursive, Forecaster Direct, LGBM Regressor, Linear Regression. The accuracy of the models was assessed by 4 indicators: mean absolute error (MAE), mean relative error (MAPE), root mean square error (RMSE), and correlation coefficient (R square or R2).
