Loading...

Developing a Machine Learning–Based Tolerance Design Approach to Ensure Reliable Performance of Mechanical Systems

Hassani, Hossein | 2025

6 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 58903 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Khodaygan, Saeed
  7. Abstract:
  8. In mechanical design and manufacturing engineering, tolerance design is pivotal in managing dimensional and geometrical deviations that arise from manufacturing and measurement errors. During the tolerance design process, dimensional and geometrical tolerances are specified and allocated, taking into account factors such as manufacturing costs, assembly quality, and reliability. However, most existing methods primarily focus on manufacturing costs and quality requirements, often neglecting assembly reliability. The allocated tolerances can significantly influence the performance reliability of mechanical assemblies throughout their specified service life. In this regard, the present study introduces a comprehensive multi-objective, time-dependent reliability-based tolerance design optimization (TRBTDO) approach. This innovative method integrates advanced techniques including machine learning, Non-dominated Sorting Genetic Algorithm II (NSGA-II), dimension reduction methods for tolerance analysis, Monte Carlo simulations (MCS) for reliability analysis, and a Shannon entropy-based Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). The approach initially reformulates the tolerance allocation problem as a time-dependent reliability-based bi-objective optimization problem, simultaneously considering total manufacturing costs and quality losses. To ensure consistent product performance throughout its specified service life, time-dependent reliability is incorporated into the model. The NSGA-II and MCS reliability methods are then employed to solve the optimization problem and generate a set of non-dominated solutions for tolerance allocation. Lastly, the enhanced Shannon entropy-based TOPSIS method is used to automatically select the optimal solution from the Pareto front. To reduce the computational burden, a Bayesian Long Short-Term Memory (LSTM) network serves as a surrogate model for the performance function, while the Univariate Dimension Reduction (UDR) method is applied as an efficient and sensitivity-free approach to tolerance analysis. The effectiveness of the proposed approach is demonstrated through a case study on a turbine clearance control system, an engineering assembly, to achieve optimal and reliable tolerance allocation
  9. Keywords:
  10. Tolerance Design ; Tolerance Allocation ; Tolerance Analysis ; Reliability Based Design ; Machine Learning ; Bayesian Long Short Term Memory (LSTM) ; Univariate Dimension Reduction Method

 Digital Object List

 Bookmark

No TOC