ProductionNewPilot Pilot, prototype, early rollout or time-limited trial; results are limited or not yet measured.

AI in the paint shop: predicts dust defects from booth sensors and guides robots to sand and polish each flaw (BMW, Audi)

Added 2026-10-30 · Last updated 2026-10-30

Manufacturers in this case: Audi · BMW · all manufacturers using AI

Role

Paint and coating lines (wet paint, e-coat, powder coating) and finishing · Typical users: paint shop or coating line supervisor, process engineer, quality engineer, finishing team lead

The problem

On a paint or coating line, a speck of dust, a bump or a run is found only after the part is painted, and then it must be sanded and polished by hand, repainted or scrapped. The causes (dust in the incoming air, filters, booth humidity and temperature, bath chemistry, cleaning stations) change from day to day, and the link between process conditions and defects is hard to see. At BMW's Munich plant, despite modern filtration, the amount of fine dust in the paint lines varies with the ambient air drawn in; above a threshold, the still-wet paint traps particles. At BMW's Regensburg plant, up to 1,000 vehicles a day go through finishing, and the small specks and bumps that appear after the topcoat are in different places on every body, while robots are normally programmed to repeat the same path. At Audi's Neckarsulm site, faults in pretreatment and cathodic dip painting were found late, causing manual work and follow-up costs.

How AI helps

Three uses of AI around the paint line. First, prediction: at BMW Munich, data from the automatic surface inspection of every painted body was built into a database, and AI compares live readings from dust particle sensors in the paint booths and dryers with it; the algorithm watches over 160 features of each body to predict paint quality, spots seasonal dust trends early enough to bring forward a filter change, and pointed out that the ostrich-feather dust-removal station needed fine-tuning. Second, finishing: at BMW Regensburg, cameras and deflectometry (striped light patterns reflected on the paint) record the exact position, shape and size of every deviation, an algorithm trained on the knowledge of the best finishers decides which ones need work, AI-controlled robots sand and polish only those spots on each body, and a laser projector marks the remaining areas (edges, joints, the fuel flap) for trained employees. Third, process monitoring: at Audi Neckarsulm, its own AI system ProcessGuardAIn watches machine and sensor data in real time to detect anomalies in cathodic dip painting and optimise chemical dosing in pretreatment, and informs the specialists; a separate AI controls dryer temperature and air volume to save energy.

Step by step

Start a defect log if you do not have one: for every part or batch, the defect type (dust, run, orange peel, crater, thin film), where it is on the part, and the date, shift and line.
Collect the process data you already have for the same periods: booth temperature and humidity, filter change dates, oven or dryer temperature, line speed, bath or powder batch, and cleaning schedules.
Look for patterns between the two (season, shift, days after a filter change, a batch) before buying anything; a few weeks of good data often shows the main causes.
If the patterns are real, test a model or a vendor tool on past data first, then run it in parallel with your current checks and log each warning and what was found.
Turn warnings into actions people can take (change a filter earlier, adjust humidity, clean a station) and record whether the defect rate moves.
Measure defect rate, rework hours and repaints before and after, and keep the final visual inspection by trained people.

Tools

BMW's own AI dust particle analysis (sensors in paint booths and dryers compared with surface inspection data): piloted at BMW Munich
Automated surface inspection (deflectometry), AI-controlled sanding and polishing robots and laser marking: used at BMW Regensburg (supplier not named in the source)
ProcessGuardAIn, Audi's own AI process monitoring, and an AI dryer control developed by Audi, the appliedAI initiative and CVET GmbH: piloted at Audi Neckarsulm
Other tools to evaluate: automated paint and surface inspection systems, the anomaly detection or process analytics module of your MES or historian, a statistics tool on your line data
For a first step without new software: your defect log and booth or oven data in a spreadsheet, plus any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below

Result

BMW Group, Munich plant: AI dust particle analysis in a pilot (BMW press release, May 2020); it predicts paint quality 'very accurately' from over 160 features, according to BMW, which filed several patents; series use was planned once more measuring points and sensor data were available. No figures published. BMW Group, Regensburg plant: Automated Surface Processing with AI-controlled robots in series production since March 2022, after automated surface inspection and before automated laser marking; described as the first car plant to run this three-step process in standard production, with up to 1,000 individually processed vehicles a day, more stable operations and shorter lead times, and being rolled out to other plants (BMW press release, May 2023). No savings or defect figures published. Audi, Neckarsulm paint shop: anomaly detection in cathodic dip painting and dosage optimisation in pretreatment in a pilot phase, series use planned for the second quarter of 2026; AI dryer control tested in series production, with the energy saving to be measured by summer 2026 (trade press, February 2026). No figures published yet.

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

All three sources are company press releases or trade press based on them; none publishes a defect rate, saving or before-and-after figure. BMW Munich and Audi's main uses were pilots at the time of writing.
These are large car plants with automatic surface inspection on every body; a smaller coating line can start with a defect log and booth data, which already show many causes.
A model that predicts defects only helps if someone acts on the warning; agree in advance who changes the filter, adjusts the booth or stops the line.
Keep trained people for the final inspection and for areas robots cannot reach (edges, joints, fragile parts), as BMW does.

Try it yourself

You are a process engineer on a paint or coating line.

I will paste: (1) our defect log for the last [3 to 6] months [date, shift, line, part or batch, defect type, location on the part, rework or scrap]; (2) process data for the same period [date and time, booth temperature and humidity, oven or dryer temperature, line speed, filter change dates, bath or powder batch, cleaning or maintenance events].

Do this:
1. Summarise the defect log by type, line and shift, and tell me which spreadsheet formulas or pivot tables to use for the counts and rates; do not compute totals or percentages yourself.
2. List possible links between defects and process conditions (season, humidity range, days since filter change, batch, shift, after maintenance) and, for each, quote the rows that support it and say how strong the evidence is.
3. Propose 3 to 5 checks or small trials to confirm or reject each link, with what to measure and for how long.
4. Draft a one-page daily checklist for the line based on the strongest links.
5. List the data we are missing to go further and how we could start recording it.

Do not claim a cause is proven: label each one as a hypothesis to test. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Want more like this? Join the newsletter

Related cases

All 13 Production use cases →