Add a real-time inspection layer to every part on the test station with Roboflow vision AI for pressure test leak detection.
Find the leak bubble in a tank full of bubbles:
- Detect leak bubble streams during immersion and dunk tank pressure tests with an RF-DETR model trained on your own tank, lighting, and parts, so the model learns the size, rate, and trajectory of a leak stream versus the ambient bubbles that rise from fixtures, part surfaces, and fill turbulence.
- Track bubbles across frames to confirm a persistent stream from a fixed origin, which is the signature of a leak, rather than a one-off bubble shedding off the part.
- Start from a small dataset: one pipe manufacturer trained a working model on about 50 labeled frames from their test tank and saw it separate leak bubbles from background bubbles on the first pass.
Locate the leak and extend to dye and soap tests:
- Map the bubble stream origin back to the part to report where the leak is (weld seam, thread, socket joint, casting wall) so the part goes to the right rework station with a photo instead of a note.
- Detect dye penetrant bleed-out and soap-solution bubble formation on tanks, vessels, and large fittings that do not fit in a tank.
- Cover pipe, fittings, valves, castings, HVAC and refrigeration components, and medical device housings with the same Workflow and separate acceptance thresholds per part family.
Write pass/fail to the station and keep the record:
- Send pass/fail, leak location, and a timestamped video clip to the test station PLC or test controller over OPC UA, MQTT, or REST so a fail holds the part and a pass releases it.
- Run alongside pressure decay and helium mass spec instruments as a visual confirmation layer: the instrument says a leak exists, the camera shows where.
- Log every test with the part serial, test pressure, hold time, model version, and clip so quality can answer a customer or certification body with the record.
Bring intelligence to every pressure test today. Stop pinhole leaks and porous castings from becoming field failures and lot recalls.
Frequently asked questions
What is pressure test leak detection with Vision AI?
Pressure test leak detection with Vision AI uses a camera on the test station and a trained computer vision model to find the bubble stream, dye bleed-out, or soap-solution bubble that indicates a leak while a part is under pressure in an immersion tank, on a hydrostatic or pneumatic bench, or in a dye or soap test cell. The model is trained on your own tank, lighting, and part families, learns to separate leak bubbles from the ambient bubbles always present in the tank, and reports the leak location along with a pass/fail written to the test controller. Every test is logged with a clip and the test parameters so results can be documented under ASTM E515, ASTM F2096, ASME BPVC Section V Article 10, and ISO 20485.
Can Vision AI tell a leak bubble apart from the bubbles that are always in the tank?
Yes. A deep-learning detection model trained on frames from your own tank learns the difference: a leak produces a persistent stream of similar-sized bubbles from a fixed point on the part, while ambient bubbles are larger, irregular, and shed once. Tracking across frames adds a time dimension so a single stray bubble does not trip a fail.
Does pressure test leak detection support ASTM E515 and ASME BPVC Section V Article 10?
Yes. ASTM E515 (the standard practice for leaks using bubble emission techniques, covering immersion and liquid-film methods) and ASME BPVC Section V Article 10 (the leak testing article of the Boiler and Pressure Vessel Code, which references bubble test procedures) define the test method, hold time, and acceptance criteria. ASTM F2096 covers gross leaks in medical packaging by internal pressurization, ISO 20485 covers tracer gas methods, and ASTM D1599 covers short-time hydraulic failure pressure for plastic pipe. Roboflow serves as the inspection engine: it watches the test, records the indication, and returns a result with the clip. Your quality and NDT teams own the written procedure, the acceptance criteria, and the qualification of the method.
Can it integrate with our test station, PLC, and quality system?
Yes. Roboflow writes pass/fail, leak location, and confidence to a test station PLC or test controller over OPC UA, MQTT, or REST, so a fail can hold the part in the fixture and a pass can release it, and the same result can flow to pressure decay or helium mass spec test software from Uson, Cincinnati Test Systems, or ATEQ as a visual confirmation record. Results, clips, and test parameters can be written to a database or pushed to quality and MES platforms such as Ignition, AVEVA, SAP, and Sparta TrackWise. Every decision is stored with the part serial, timestamp, and model version, which gives quality the traceability a customer audit, NSF listing review, or 8D investigation asks for.
