Books
One family takes an engineering discipline’s own questions and follows them through to a model, a computation and a decision. A second teaches an application domain from no assumed background, one chapter per lesson. The third states the method itself — the structure of a physical system is also the structure of every probabilistic question one can ask about it — and carries it through a general text and five domain editions. All free to read, and every number is generated from a committed result.
Working drafts, revised as the underlying studies land.
Each of these takes one engineering discipline’s own questions — asked in the words a practitioner uses — and follows them through to a probabilistic model over a physical system, its derivation, the computation and the decision it supports. One framework and one notation throughout, with a plant that grows as the book does.
Turning a reliability question asked in ordinary language into a probability or expectation over the unknowns of a physical model: the plant graph, the question, the evidence, the derivation and the checks that decide whether the answer can be trusted.
Failure and degradation written as physical models, competing explanations ranked against the evidence, risk assembled from the parts that cause it, and the acceptance tests an answer has to pass before it is used.
Maintenance, spares, inspection and investment as decisions computed from the posterior, and worked cases that carry a question from the plant graph through to the choice it informs.
Virtual sensors for temperatures nobody measures, heat flows inferred rather than metered, inconsistent meters reconciled, operating envelopes as joint limits, and faults isolated from residuals — the questions a control room asks during a shift.
Set-point changes, outages and load transfers posed as scenarios, look-ahead risk with no intervention, and the goodness-of-fit checks that say whether the model still describes the plant it is advising on.
The asset manager's questions over the life of an asset. A health index is a parameter of the physics — a coil's conductance, a battery's capacity — reported against its commissioned value, with load and weather entering as known inputs. Then the rate that parameter moves at, whether it is accelerating, and whether a step has occurred, separated from the season.
Remaining useful life and the chance of finishing a mission, repair or replace, when to inspect and what trigger to set, which asset to attend to first, and what a condition monitor is worth before it is bought.
Will the plant carry the load, and how much to build. Load-carrying capability under a design day, growth and diversity in the load path, and redundancy and sizing decided from the posterior rather than from a rule of thumb.
Where the binding constraint is and what it is worth moving, how the margin splits across a shared path, and what the uncertainty in each design input costs — which turns into which measurement to buy before committing. Then the design with a unit out, the cascade that can follow, and what the remaining risk is worth in money.
The state is a posterior and the controller is a decision over it. The closed loop written as a model with the controller inside it, what stability means for the fit, and the belief the controller acts on — the Kalman recursion as conditioning, and what the loop leaves unobservable.
Where certainty equivalence holds and the covariance drops out of the best move, integral action against a bias the model does not carry, and tuning read off the posterior. Then the move under a limit, over a horizon, against a forecast that is itself uncertain.
Written to teach rather than to state a method: each book starts from no assumed background in its subject and ends where the modeling questions begin. Each stands alone.
Taking a question asked in ordinary language — in a meeting, by an operator, by a reliability engineer — and writing down the probabilistic problem it actually asks, through one seven-step method and a one-page formulation sheet. It stops where the solving starts, and it says when a question cannot be answered as asked.
What is forecastable, baselines, features, models, validation, probabilistic forecasts, fleet forecasting, and the operational decisions the forecast is for.
Failure definitions, lifetime and hazard models, system logic from block diagrams to cut sets and importance measures, dependent failures and probabilistic risk assessment, and repairable systems and availability.
Degradation and prognostics, power-system adequacy, observability, state estimation and bad-data detection, and the engineering decisions reliability analysis is for — maintenance, spares, investment and the value of information.
Concepts and terminology behind the wind at 10 to 150 metres: pressure forcing, the boundary layer, terrain, measurements and numerical models.
From DER fundamentals to a worked virtual power plant: demand response, DERMS, thermostats, vehicles and batteries, fleet aggregation, dispatch, load shaping and grid services.
Data centers as physical systems: power, heat, cooling, measurement, control, efficiency, reliability, diagnosis, capacity, optimization and grid flexibility.
A tutorial for a reader with no battery background: the machine, electrical quantities, the cell, the circuit model, the hardware, heat, aging, faults, and what a telemetry record can determine.
One idea, followed through: the structure of a physical system — where conserved quantities accumulate, how they move, where they cross the boundary — is also the structure of every probabilistic question one can ask about it. A general text states the method; five domain editions teach the same method with that domain’s own examples and case studies.
The method in its general form: conservation graphs, probability attached to the physics, inference, the operator's questions, time, hierarchy and scale.
A rack's inlet-temperature alarm trips: recirculation, a fouled coil, a misplaced sensor, or a load above its nameplate — and can the row take the next deployment?
The state estimator flags a residual at the morning ramp: a failed meter, a wrong breaker status, a drifted line, or a real flow?
A district's night flow climbs: a burst, a drifting meter, a boundary valve that is passing, or legitimate use — and where should the crew go?
The return temperature creeps up: a fouled exchanger, an open bypass, a biased sensor, or real load — and at which substation?
A rack drops out forty minutes into a two-hour discharge: a faded cell, a risen resistance, a loose connector, or a biased sensor — and can the site still meet the hour it sold?
