Commercial Solutions
Computer vision engineering for manufacturing
From evaluating whether an inspection is technically viable through to a system running on the line.
Start with feasibility
Not every inspection should be automated, and the cheapest time to find that out is before a production system is designed around it. Feasibility work answers a narrow, practical question first: can this defect or measurement be detected reliably, on these parts, under conditions the line can actually maintain?
When the answer is yes, the study produces the imaging and processing basis for the production system. When it is no, or not yet, that is a useful result too.
Service 01
Vision Feasibility Studies
Evaluate whether a proposed inspection can be reliably automated before committing to a full production system.
Many inspection ideas fail for reasons that show up early: the defect is not visible under the available lighting, the required resolution is impractical at line speed, or the acceptance criteria are inconsistent between operators. A feasibility study surfaces those issues while they are still inexpensive to address.
The work centres on imaging the actual parts. Sample images are collected under candidate lighting and optics, analysed against the defect or measurement of interest, and used to judge whether the approach can be made repeatable in production.
- Review of the inspection requirement and acceptance criteria
- Lighting, optics, camera, and resolution assessment
- Sample image collection and analysis on representative parts
- Identification of the conditions that would break the inspection
- A written technical recommendation, including when the answer is no
Service 02
AI Defect Detection
Use computer vision and deep learning to automate visual inspection and identify manufacturing defects.
Some defects are well described by rules: dimensions, presence and absence, edges, and thresholds. Others are easier to demonstrate with examples than to specify, and that is where trained models are useful. Both approaches have a place, and the right one depends on the defect, the sample data available, and how the result will be used.
Work covers the whole path: defining classes with the people who make the accept and reject decision, organising image data, training and evaluating models, and deciding how results are reported to operators and to the control system.
- Classical image processing and deep learning approaches
- Defect classification, detection, and segmentation
- Image dataset collection, labelling strategy, and review
- Model evaluation against false accept and false reject cost
- Inline and offline inspection workflows
Service 03
Custom Vision Engineering
Industrial camera integration, image processing, Python/OpenCV development, GPU inference, PLC integration, and production deployment.
Production vision systems are rarely a single algorithm. They involve cameras and lighting that must be selected and mounted, triggering that must line up with the process, software that must run unattended, and results that must reach the control system in a form it can act on.
Custom engineering covers the pieces between those parts, including integration with existing equipment, so a vision system does not have to be built from scratch when a line already has usable hardware.
- Python and OpenCV application development
- Industrial camera, lens, and lighting integration
- Trigger, timing, and image acquisition pipelines
- GPU inference and runtime optimisation
- Edge deployment on industrial hardware
- PLC and industrial system integration
Capabilities
Technical areas we work in
Individual capabilities that make up the services above, available as part of a larger project or as focused support on an existing system.
Industrial camera integration
Area-scan and line-scan cameras, GigE Vision and USB3 Vision interfaces, lens and lighting selection, triggering, and mounting considerations.
Python / OpenCV development
Image processing pipelines, measurement and alignment routines, calibration, and tooling built on standard, well-supported libraries.
GPU inference and optimisation
Model runtime selection, batching and precision choices, and profiling to meet the cycle time an inspection actually has available.
Edge AI deployment
Running inspection on industrial PCs or embedded hardware near the line, with attention to startup behaviour, logging, and recovery.
PLC and automation integration
Passing inspection results to the control system for reject handling, and coordinating vision with the sequence the machine already runs.
Production deployment support
Commissioning, image logging for later review, and documentation so plant personnel can maintain the system after handover.
Every application differs in parts, line speed, and existing equipment, so scope is defined per project after reviewing the inspection requirement.
Not sure if your inspection is a vision problem?
Describe the part and what needs to be checked, and we will tell you what would be needed to evaluate it.