To improve the performance and economy of a manufacturing system, the Prognostics and Health Management (PHM) and Remaining Useful Life (RUL) of its vital components are attracting attentions from scholars. As one of the most important components, the RUL of a machining tool determines the performance of a manufacturing system and product quality. In addition to physics-based and data-driven models, the hybrid model for RUL prediction is attracting interests recently, especially the hybrid physics-based data-driven model, which integrates the physics-based method and data-driven method to unify advantages of them.

The physics-based part of hybrid model is based on physical mechanism of the process and product and could provide the non-geometric dimensioning interior features of products, such as residual stress, which may determine the workpiece material properties and the product performance. As a product is composed of its profile and material, such interior features should be fused into manufacture quality control and the tool RUL prediction. A novel method of monitoring workpiece quality considering interior features is presented. This method provides simulation results of workpiece’s non-geometric dimensioning interior features based on process scenario data. By extracting data from off-line database according to the process scenario and creating an on-line simulation model, the proposed rapid Finite Element Method can complete the simulation of workpiece interior features rapidly, and the simulation results could supplement classical production quality control. This algorithm is proposed and discussed mathematically based on the multi-subdomain coupling method, and the simulation error is estimated. A case study of a rolling production process shows that this method is effective and could be used to monitor workpiece quality in-process.

Specifically, as one major factor of processing scenario, the wearing of tool reduces its capability of production yield and diminishes product quality. Therefore, a capability-based RUL prediction approach is proposed to thoroughly evaluate the state and RUL of machining tools. First, the connotation of the capability of tool is discussed, and a framework for quality assurance capability-based RUL prediction is proposed. Product quality, which can be used to assess the capability of tool, is modelled and expanded in consideration of non-geometric dimensioning and tolerancing (non-GD&T) features based on the classic geometric dimensioning and tolerancing (GD&T) system. Second, a physics-based model of the process is developed to estimate the non-GD&T features and calculate tool wear. Third, a hybrid data-driven and physics-based model is proposed to quantitatively assess the capability of tool. Finally, a case study of a rolling machining tool is carried out to verify the effectiveness and proactiveness of the proposed approach, the final result highlights its accuracy in estimating the RUL of machining tools with better interpretation.

Under the real processing scenario, noises and uncertainties are unavoidable. their influence on tool wear and RUL could be well estimated by data-driven method. However, the data-driven model of tool RUL is usually based on wear amount or a limited number of indices and may not directly handle the details of tool wear well, such as geometry changes. Therefore, a hybrid model is proposed to estimate the tool RUL based on the tool details and process uncertainties. First, a physics-based model of process and tool wear is built, which provides exhaustive data of process, including the situations and interactions between components, and tool wearing. A basic group data of tool wear details is obtained via several Monte Carlo (MC) simulations of this physics-based model. Second, the detailed data is decomposed by PCA and several non-linear Wiener process models are fitted based on the decomposed data to estimated uncertainties. Third, by reconstructing prediction data from non-linear Wiener process models, the tool wear details with uncertainties could be calculated, and based on this data, the tool RUL is estimated. Finally, a case study is developed to illustrate the effectiveness.