Research

Industrial Relevance

My research addresses problems that arise directly in product qualification, reliability demonstration, and life-testing in high-tech manufacturing. Specifically, I develop methods to design accelerated life tests (ALT) that extract maximum statistical information about product lifetime under budget and time constraints; I build inference procedures for Weibull and related lifetime models from small, censored, or interval-monitored samples, including settings with multiple competing failure modes and heterogeneous failure populations; and I construct Bayesian frameworks for reliability demonstration testing that quantify confidence in qualification decisions under uncertainty. These methods are applicable wherever product testing is expensive, field data are scarce, and lifetime prediction is safety-critical or economically essential, including energy storage systems, medical and pharmaceutical devices, environmental control equipment, and safety-critical system certification.

Methodologically, my work spans frequentist, Bayesian, and robust inference, optimal experimental design, and uncertainty quantification for lifetime characteristics.

Research Areas

1. Inference and Uncertainty Quantification for Life-Testing Models

Lifetime testing plays a key role in reliability engineering, where the goal is to understand failure mechanisms and predict product lifetimes under limited testing resources. I develop frequentist, Bayesian, and robust inferential methods for censored and accelerated life-testing experiments, including step-stress, cyclic-stress, and heterogeneous models, focusing on realistic experimental constraints such as interval monitoring, competing failure modes, and heterogeneous failure populations. My work integrates uncertainty quantification into inference, formally characterising uncertainty in lifetime model parameters and quantiles, and addresses situations involving limited testing time, budget constraints, and contaminated data, where efficient and robust use of data is essential. These methods are applicable wherever components undergo repeated or elevated stress in service and where reliable lifetime prediction is safety-critical or economically essential; such as energy storage systems (batteries, capacitors), environmental control equipment (air conditioners, HVAC systems), medical and pharmaceutical devices, and protective gear, among others.

2. Optimal Decision-Making and Experimental Design

A second strand of my research concerns optimal decision-making under uncertainty in life-testing and statistical quality control. On the design side, I study the optimal planning of life-testing experiments with the aim of maximising inferential efficiency or minimising overall testing cost; choosing stress levels, censoring schemes, and sample sizes so that the experiment extracts maximum information about lifetime model parameters under real budget and time constraints. I am also developing computationally tractable Bayesian optimal design via Gaussian process surrogate modelling, replacing costly repeated posterior evaluations with a trained surrogate to reduce computational cost while preserving inferential validity. On the decision side, I develop Bayesian acceptance sampling plans in statistical quality control: using decision theory, I design accept and reject rules that balance inspection cost, product quality, and the risks faced by both producers and consumers. These methods are motivated by applications in manufacturing and pharmaceutical quality control, where decisions about product quality must be statistically sound.

3. Inference for non-regular families of distributions

A part of my research concerns statistical inference for non-regular families of distributions, which do not satisfy the standard regularity conditions assumed in classical statistical inference. In such models, the support of the distribution may depend on unknown parameters (for example, Uniform(0, θ) or three-parameter gamma models), likelihood functions may be unbounded or non-differentiable, and Fisher information may be infinite or undefined. As a result, classical properties such as the existence, consistency, and asymptotic normality of maximum likelihood estimators may fail. I am interested in developing specialized inferential methods for these settings, with particular emphasis on estimation and model selection.

Computation and implementation

My research places strong emphasis on computational feasibility. I implement my methods using modern statistical computing tools, including R and Bayesian frameworks such as Stan, and make use of high-performance computing environments when required to scale algorithms to complex models and large simulation studies.

Broader perspective

My background enables interdisciplinary collaboration at the interface of statistics, reliability engineering, and actuarial science. I am particularly interested in problems where rigorous statistical theory, computation, uncertainty quantification, and real-world decision-making intersect in industrial and applied settings.