When the Inspection Workbook Has Three Editors and No Reproducible Math

The chemical plant access window is short, frequent and announced late

A batch or specialty chemical site does not run on a turnaround cycle in the petrochemical sense. It runs campaigns. A reactor train makes one product for six weeks, comes down for cleaning and changeover, and starts the next. A dryer is idle between orders. A tank farm rotates service. Each of these creates an access window measured in days, and the windows are frequently confirmed with two or three weeks of notice because they follow the order book, not an engineering plan. That is a fundamentally different scheduling problem from planning a single shutdown eighteen months out.

The consequence is that a due date, on its own, is nearly useless. What the inspection group needs at any moment is a ranked queue: here is the work that is ready to execute, here is what each item needs in the way of access, crew and equipment, and here is how long it takes. When operations announces that Reactor 4 comes down on the twelfth for nine days, the answer to what goes in that window has to be available the same afternoon, not after a week of workbook archaeology. Queue readiness, not date arithmetic, is the differentiating capability.

A spreadsheet cannot represent readiness because it has nowhere to put the prerequisites. Access condition, required certification, calibration status of the instrument, whether the vessel needs to be cleaned and gas freed first, whether a scaffold or man-way removal is required - each is an attribute of the task, and each determines whether that task can go into a nine-day window. Once you start adding those columns to a workbook you are building a database with the wrong tool, and the formulas that stitch them together are exactly the ones that break when someone sorts the sheet.

A mixed asset population defeats a single calculation model

The chemical asset base is broader and more heterogeneous than most fixed-equipment software assumes. Carbon steel vessels sit alongside glass-lined reactors, alloy 20 and Hastelloy columns, PTFE and rubber lined piping, filament-wound FRP and dual-laminate tanks, graphite heat exchangers and jacketed kettles. Each of those degrades differently and is inspected differently, and only some of them have a thickness that means anything. A single corrosion-rate column applied across the whole population is not a simplification; it is a category error that produces confident, wrong numbers for half the plant.

Glass-lined equipment is the clearest example. Its condition is assessed by spark testing at the manufacturer's recommended voltage and by close visual examination for chips, crazing and spalling, with the interval driven by thermal shock history, mechanical handling and the severity of the service rather than by any measurable wall loss. Reinforced thermoset equipment is another: ASME RTP-1 governs its construction, and visual acceptance of laminate defects follows the classification scheme in ASTM D2563, which has nothing to do with ultrasonic thickness at all. Lined piping fails at joints and through permeation, not through general thinning.

A scheduling module for this industry therefore needs several interval models coexisting: thickness-based with a computed remaining life for the steel equipment, condition-based for linings and non-metallics, time-based for devices tested on a fixed cycle, and risk-based where an assessment has been done. Critically, each asset must declare which model governs it, so that a report can be filtered by model and reviewed. The moment a lined vessel appears in a thickness report with a corrosion rate of zero and infinite remaining life, someone will believe it.

Pressure relief devices are where task volume actually lives

Count the relief devices on a multi-product chemical site and the number usually dwarfs the vessel count several times over. Each one has a test interval, a set pressure, a service classification and a bench test record, and each one has to come off, go to a shop, be tested, repaired or replaced, and be reinstalled with the right nameplate data. API 510 sets maximum test intervals by service severity, with the longest cycles reserved for clean, non-corrosive service and the shortest for fouling or corrosive duty, and API 576 covers the inspection practice itself. This is the highest-volume recurring task set in the plant.

Batch operation complicates it in a way continuous plants do not face. The service a relief device sees depends on which product is running. A device that is in clean service during one campaign may be in fouling or polymerising service during another, and the interval that was reasonable a year ago may not be reasonable now. Attaching the interval to a fixed assumption made when the register was first built is how a device ends up stuck open on the day it matters. The interval basis has to be attached to service condition, and a change of product has to be a trigger to review it.

Spares management is the operational half of this, and it is where a spreadsheet gives up entirely. To pull a device for testing you need an installable spare of the right set pressure and materials, which means the schedule and the inventory have to know about each other. Building inspection scheduling on an ERP rather than a standalone tool is worth doing largely for this reason: the same system that raises the test task can reserve the spare, raise the purchase requisition when stock is short, and keep the certification paperwork attached to the serial number.

Reproducibility is a regulatory requirement, not a matter of taste

The reason a non-reproducible calculation is a serious problem, rather than an annoyance, is that OSHA 29 CFR 1910.119(j) requires documentation of each inspection and test that identifies the date it was performed, who performed it, the identifier of the equipment, a description of the inspection or test, and the results. A due date that nobody can derive from those records is not documentation; it is an assertion. When a compliance audit or an insurer's engineering survey asks how a particular interval was established, pointing at a cell in a workbook that has been edited by three people over four years is not an answer.

There is a second regulatory thread specific to chemical manufacturing. Many small ASME-stamped vessels in this industry sit under state jurisdictional boiler and pressure vessel programmes, with in-service inspection and any repair or alteration performed in accordance with the National Board Inspection Code, NB-23, by an organisation holding the appropriate certificate of authorisation. That means a repair record has to link to the vessel's national board number and to the authorised inspector who accepted it. Workbooks almost never carry those links, and reconstructing them years later is painful.

Reproducibility in practice means three things. Stored inputs, not stored outputs, so that a remaining life is recomputed from readings every time it is displayed. An immutable change log naming the user, the timestamp, the field, the old value and the new value. And a required basis for any manual override of a system-derived date, so that when someone extends an interval, the reason is attached to the extension rather than living in an email. None of that is exotic; all of it is impossible in a shared workbook.

The specific ways your current workbook is already wrong

Start with the sort. The most common structure in an inspection workbook is one row per CML with the previous reading in one column and the current in the next, and a corrosion rate formula between them. Some of those formulas use relative references and some use absolute, because they were written by different people at different times. The first time somebody sorts by due date to prepare for a planning meeting, a subset of the formulas keep pointing at their original rows. The rates change. Nothing turns red. The sheet is now internally inconsistent and there is no way to detect it short of recomputing everything by hand.

Next, look for pasted values. Somebody had a circular reference or a display problem and fixed it by copying a column and pasting as values. The numbers survived; the derivation did not. From that point the column is a set of historical constants that no longer respond to new readings, and the next person to add a survey will enter the readings and wonder why the rate did not move. In a workbook that has passed through three owners, expect to find at least one such column, and expect it to be one that feeds due dates.

Then check the dates. Inspection data frequently arrives from contractors as CSV, and a CSV date in an unexpected format imports as text. A text date compared against a real date returns a value that neither errors nor matches, so the conditional formatting that highlights overdue items simply skips those rows. Assets vanish from the overdue list without vanishing from the sheet. This one is worth checking today, before any software decision: filter for cells that are not date-typed and see how many assets have been invisible.

Getting off the workbook without a bad first ninety days

Do not begin by cleaning the spreadsheet. The workbook is your only record of what happened, and editing it to make it importable destroys the evidence of its own inconsistencies. Import it exactly as it is into a staging area, then let the destination system classify every row: parsed cleanly, parsed with a warning, or uninterpretable. What comes back is a defect list of your own data, ordered by severity, and it is the most valuable document produced by the entire project. Work it with an engineer who has the authority to make a judgement, and record every decision as it is made.

Sequence the cutover by asset class rather than by area. Move the thickness-based steel equipment first, because it is the population where the destination's calculation can be verified against the workbook's answer directly - if the new system reproduces the old remaining life on that population, the arithmetic is trustworthy. Then bring across the relief device register, which is high volume but structurally simple. Leave the lined, non-metallic and condition-based assets for last, because those need an interval model defined before they can be loaded, and defining it properly is engineering work, not data entry.

Run the two in parallel for one cycle, with the workbook read-only. The temptation to skip this is strongest in a small inspection group, which is exactly where the parallel run pays most, because a small group has no redundancy if the schedule turns out to be wrong. Atlantis implements this on Odoo, so inspection scheduling shares records with maintenance work orders, spares, technician certification and calibration rather than sitting beside them, and the migration is delivered as a defined project with a reconciliation report as the acceptance criterion. Ask for a consultation at info@atlantisndt.com.

Request a consultation