From Method-Centric Optimisation to Failure-Oriented Robustness: A Framework for Industrial Inspection Systems

Robust inspection of aeroengine turbine blades remains a critical challenge in safety-critical industrial environments, where limited data availability, class imbalance, optimisation bias, and imaging degradation can reduce the reliability of learning-based inspection systems. Existing studies mainly improve robustness through isolated algorithmic optimisation, while providing limited understanding of why inspection failures occur under different industrial conditions. This study proposes a Failure-Oriented Robustness Framework that categorises robustness degradation into three dominant failure sources: distribution failure, decision failure, and representation failure. The framework establishes a systematic relationship between failure diagnosis, intervention selection, and robustness interpretation. The applicability of the proposed framework is demonstrated through representative turbine blade inspection scenarios in-volving two-dimensional surface defect detection and three-dimensional CT image enhancement. Data-level experiments investigate augmentation behaviour under different dataset characteristics, model-level experiments analyse optimisation objectives and architectural factors, and imaging-level experiments evaluate CT enhancement under degradation conditions. Experimental results demonstrate that robustness improvement depends on the alignment between intervention strategies and dominant failure sources rather than algorithmic complexity alone. Under severe distribution failure, targeted augmentation improved mAP@0.5 from 0.290 to 0.599, while objective alignment increased detection performance from 0.588 to 0.710 without increasing model complexity. For CT representation recovery, the proposed enhancement approach achieved a 3.21 dB PSNR improvement compared with U-Net. The findings demonstrate that robustness is a failure-dependent property rather than an intrinsic characteristic of individual methods. The proposed framework provides an interpretable analytical perspective for diagnosing robustness degradation and guiding failure-oriented intervention design in safety-critical industrial inspection applications.

@article{Zhou2026,
  title     = {From Method-Centric Optimisation to Failure-Oriented Robustness: A Framework for Industrial Inspection Systems},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Ying Zhou and Mei Choo Ang and Ummul Hanan Mohamad and Ah-Lian Kor and Kok Weng Ng and Saraswathy Shamini Gunasekaran and Halimah Badioze Zaman},
  doi       = {10.14569/IJACSA.2026.0170773},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170773}
}