The Mantua Group

Simple Black and White Asset Management, Reliability Expertise, and Maintenance Execution Perfection.

Data Analysis

Dealing with Data

Fitness for Purpose, Censoring, and the Discipline of Turning Reliability Data into Reliability Decisions

A reliability engineer who has worked in this field for any length of time has been handed a stack of data and asked to make sense of it. The stack is rarely fit for the question that motivates the engagement. The collection system was usually designed for a different purpose by people in a different function; the records are incomplete in ways that distort the underlying distribution, and the volume of data is sometimes large enough to disguise the absence of useful information. This paper sets out ten principles for working with reliability data well: starting from fitness for purpose, working through the structural problems of censoring and missing renewals, addressing the volume problem with reliability performance indexing, and ending with the most important and most neglected of the ten, talking to the people who collect the data so that the next dataset is better than the last one. The principles are drawn from a Speaking of Reliability conversation between Philip Sage and Fred Schenkelberg and have been translated here into a structured engineering doctrine in the TMG voice.

Quality Statistics, Reliability Statistics

Two Statistical Disciplines, Two Different Questions: How They Differ, How They Combine, and Why an Engineer Should Master Both

The two statistical traditions that reliability engineering inherits, the quality-statistics tradition descending from Shewhart and Deming and the reliability-statistics tradition descending from Weibull, Fisher, and the post-war life-testing community, are often spoken of as interchangeable extensions of a single discipline. They are not interchangeable. They were built for different questions, they make different assumptions about the data, they use different mathematics, and they support different decisions. An engineer comfortable in both moves fluently between the shop-floor control chart and the field-fleet survival analysis. An engineer who knows only one will, sooner or later, apply the wrong tradition to the wrong problem, and the analysis will produce results that are mathematically correct and engineering nonsense.

Competing Risks, Or Not?

Disciplined Failure-Mode Decomposition for Reliability Data Analysis: When the Risks Framework Applies, When it Misleads, and the Practical Questions That Tell the Difference

The decision to model failure mechanisms as competing risks or to treat them as independent processes requiring separate analyses is among the most consequential a reliability engineer makes. The choice cannot be settled by software, by professional habit, or by the convenience of an aggregated dataset. It can be settled only by a clear understanding of the physics of failure, the timing of risks exposure, and the empirical evidence that two or more mechanisms genuinely overlap on the timeline of an asset’s service life. This white paper sets out the discipline that distinguishes legitimate failure mode competing-risks modelling from analytically convenient over-aggregation, the framework that follows from each choice, and the practical questions a practitioner should pose before adopting either approach.

The Right Tool, at the Right Time, at the Right Depth

Elevating Reliability Engineering Decision-Making Through Disciplined Diagnosis, the Right Tools, Proportionate Method Selection, and Organizational Readiness

Reliability engineering offers no shortage of tools. The discipline has accumulated, across roughly a century of formal practice, an arsenal that includes Weibull life-data analysis, fault tree analysis, FMEA and FMECA, root cause analysis, control charts, modal analysis, design of experiments, accelerated life testing, condition monitoring, reliability-centred maintenance, and dozens of variants. The persistent failure mode in the field is not a shortage of tools but a mismatch between the tool and the problem: an over-engineered analysis applied to a question a five-minute hypothesis test would have settled, or a sophisticated technique deployed in an organisation that has not yet developed the foundational disciplines required to absorb its output.

Weibull Statistical Analysis for Transmission Asset Reliability

Applying Weibull for Maximum Likelihood Estimation to Left-Truncated and Right-Censored Lifetime Data

This white paper presents a rigorous statistical methodology for analyzing transmission line asset reliability using Weibull distribution analysis. The approach specifically addresses the analytical challenges inherent in utility asset data: left-truncated observations from legacy system migrations and informatively right-censored data from inspection-driven replacement programs.

A study of approximately 15,000 transmission structures across an 11,000-kilometer network demonstrates the Weibull methodology’s practical application. By employing Maximum Likelihood Estimation (MLE) rather than ordinary least squares regression, the analysis produces unbiased parameter estimates even under complex censoring conditions that would render traditional approaches unreliable.

Key findings reveal that unique classification of assets exhibit similar but different reliability characteristics, with characteristic lives (η) of 60 and 75 years, respectively, and shape parameters (β) indicating wear-out failure modes in both populations.