Structure: scoping and diagnosis

The reference method is a measurement, with an error of its own. Three faults recur: a method not specific to the quantity you are after, such as a loss on drying that captures every volatile, not only water; outliers among the reference values; samples taken away from what the measurement sees.

Examining it before committing avoids the least comfortable situation in this trade: a model that looks poor when it is the reference that is scattered. Knowing the scatter of your laboratory method tells you, in advance, the accuracy any model can reach. Why two instruments give two figures.

We do not publish an average gain: there is none. We publish the calculation method, and we apply it to your data during scoping. Five inputs are enough.

• The real durations of the operation, batch by batch, over a period long enough to contain the difficult batches.
• The hourly cost of the equipment tied up, including line occupation and not only energy.
• The number of batches per year, which decides everything. The same unit gain does not weigh the same on twelve batches and on three hundred.
• The reject and rework rate, and the cost of a reworked batch, almost always underestimated. It is spread over several departments: rework, destruction, and late delivery to the customer.
• The off-line analysis hours actually released, multiplied by the hourly cost of an analyst. It is the most often forgotten input, and the easiest to obtain: it is already in the laboratory logbook.

The recoverable margin is the dispersion of those durations, not their mean. It is the point the calculation misses most often. An operation set to a fixed duration is set on its worst case. On every batch, it carries the gap between the slowest batch and this one. Measuring makes it possible to remove that margin on the batches that did not need it. That is, on at least half of them. Conversely, an operation where all batches behave the same way has more or less nothing to gain in cycle time. The project will then be justified by quality or by release, or not at all.

The same reasoning holds on the cost side. The full cost of a measurement point goes well beyond that of the instrument. Commissioning, integration and maintenance account for the largest share over its lifetime.

When should PAT enter the capital budget? How do you defend the request to management? Will the data be usable and compliant?

A PAT roadmap connects three timelines: process, data and CAPEX. We build it with your production, quality, automation and finance teams, one measurement point at a time.

Each investment rests on a proof of concept, run at pilot scale or directly on the process, under final deployment conditions. The CAPEX request then presents results measured on your own material, an ROI estimate and a validation plan.

Data is designed in from day one: spectral data historian, control system integration, data integrity, 21 CFR Part 11 compliance.

The roadmap is reviewed every year as new products arrive, a possible move to continuous manufacturing takes shape, and the first installed points deliver feedback.

Whether the project starts from a corporate strategy or a production problem, the starting point is the same: from Process Understanding to Real-time Quality.

The technology decision first: which measurement physics answers the question, on this matrix, at this rate, with this access. We implement several, each with the partner who masters it: it is that breadth that makes the comparison possible. Scoping retains one or more candidates, and it is the proof of concept that confirms the choice of one or more PATs.

The calibration burden decision next, and it is at least as structuring. A pure-component model is built on the spectra of the pure components. It requires a known and declared composition, and discovers nothing that has not been declared. A model calibrated on samples is regressed on reference values. It requires a population of samples representative of the real variability, and therefore a campaign, and therefore line time, and therefore upkeep of the model over time. Some measurements, finally, need no multivariate model, because the quantity comes directly from the physics of the measurement. That choice decides the cost of ownership and the lead time before the first usable measurement, far more than the brand of the instrument.

These are what move schedules, and they are settled at scoping rather than during installation.

• Explosive atmospheres. Zone, equipment category, protection method. An ATEX requirement narrows the available catalogue and shapes the integration drawing, so it is asked in the first meeting.
• Cleaning and sterilisation. A wetted sensor meets clean-in-place and sometimes steam: window material, temperature rating, absence of retention. These are specification fields rather than details.
• Requalification. Drilling qualified equipment brings requalification, with its file and its downtime. It is costed at scoping, and it is frequently what makes a non-contact position preferable, optically less favourable and far lighter to put in place.
• Integration with the information system. Where the results go, who signs them, how they are kept, and what an auditor will be shown. Computerised system validation is a real gate: brought in at scoping it steers the design. What a declaration of conformity covers.

Where the project continues, scoping produces a user requirement specification. It states what the installation has to do, under which conditions, and against what it will be verified. It makes two quotations comparable, and it lets you say later whether the delivered system matches what was asked.

Its structure is stable from one project to the next.

• Contacts: who carries the need, who decides, who will run it.
• Process: continuous or batch, how it is operated, and the measurement objective in one sentence.
• Sensor: wetted material, dimensions, chemicals met, length of the connection.
• Variable factors, each with its range: temperature, pressure, flow velocity, concentration, conductivity. The block that decides the most and is filled in the least. Service conditions are declared.
• Further requirements: mechanical adaptors, ATEX, inputs and outputs, software and interfaces.
• Traceability: every requirement numbered, with its description, the technical answer proposed, the solution retained and its approval. Several signatories, you among them.

That last block is what makes the end of the project straightforward: it lets you walk through, requirement by requirement, what was asked and what was delivered.

A document that answers six questions, and that belongs to you.

• What is measurable: the quantity, and in what form: a value, a trend, a state.
• With what: the candidate technology or technologies, and the associated calibration burden.
• Where in the process: the measurement position, the installation mode, and what that position makes it possible to see or prevents from seeing.
• What it takes to prepare: mechanical access, reference campaign, line time, people to involve.
• What it would return: the calculation, with its assumptions shown, from your figures. A scoping note is not closed with an empty box where the gain should be.
• In which cases not to go ahead: the most useful part. Signal at trace level on this matrix, reference method too dispersed to calibrate, presentation variability that would dominate the signal sought, gain too small compared with the full cost. Each is a legitimate reason to stop.

If scoping concludes favourably, it leads to a trial protocol and to the acceptance criteria of the proof of concept, set before the trials. Writing the criteria after seeing the results has never demonstrated anything.