The Future of Fabric Inspection

The White Rose Lecture - Huddersfield Textile Society

The future of fabric inspection is here. That is the message Mark Shelton and David Buxton of Shelton Vision (www.sheltonvision.co.uk) gave to those attending Huddersfield Textile Society’s White Rose Lecture on the evening of 25^th November.

The well-attended lecture included members of the Society as well as those from Bradford Textile Society, The Textile Centre of Excellence and the Society of Dyers and Colourists. Held at the Textile Centre of Excellence at Outlane, Huddersfield guests enjoyed a networking session before the lecture.

From Machinery Supplier to Vision Innovator

Shelton Vision began in 1995 as part of Shelton Machinery, but following the sale of the machinery business in 1998, the company focused entirely on automated vision inspection. In 1999 WebSpector was launched on a PC platform. The first major success came in 2002 when they were selected by W. L. Gore & Associates as their supplier of automated fabric inspection. W. L. Gore are an American multinational company and many of us wear their ‘Gore-Tex’ waterproofed products.

By 2010, the automotive interior market had become the company’s largest field of application. Patterned fabric inspection was introduced at ITMA Milan in 2023, and in 2025 Shelton Vision achieved another step forward: fully integrating AI deep learning to automate defect naming, classification and grading with consistency that surpasses manual inspection.

Courtesy Shelton Vision

A System Built for Industry Reality

While each customer has different quality requirements, Shelton Vision’s approach is based on standardised modules configured for individual needs. Using high-specification industrial cameras and PC-based software, the system can be installed economically, where justified by quality, productivity, or labour demands.

The correct fabric presentation and controlled lighting remain essential, but once in operation, inspection speeds up to 300 m/min are achievable. A defect detection rate exceeding 98% is claimed.

One case study referenced automated inspection of workwear production, where the entire process required just six technicians per shift to inspect, review, cut, roll and pack 500,000 metres per week, automation delivering higher accuracy and consistent decision-making.

Five-Stage Defect Sentencing and Classification

1. Defect Detection

User-set parameters (thread density, thickness, spacing etc.) guide algorithms to detect visible deviations from expected fabric structure.

2. Debris Removal

Pre-filtering eliminates temporary artefacts such as creases, dust and lint to avoid false positives.

3. Defect Naming with AI

Automated naming and classification is now handled by a custom-engineered Convolutional Neural Network (CNN) trained on defect images. For classifying fabric defects in line-scan camera images, a CNN is well-suited because it can automatically learn the visual features that distinguish normal weave/knit patterns from faults such as slubs, broken ends, stains, or mispicks. Instead of relying on hand-crafted image processing rules, a CNN learns directly from example defect images, detecting texture changes, repetitive pattern disruptions, and subtle anomalies in the fabric structure. By sliding filters across the image, the network captures local spatial patterns and progressively builds a deeper understanding of the fabric’s texture, enabling robust and real-time fault classification even under varying lighting or production conditions.

As a result, consistent real-time classification is achieved up to 99.45% classification accuracy.

4. Defect Grading

Used during the grading and classifying stage, this image is a 2-dimensional representation of a defect using size and colour contrast, X axis shows contrast and Y-Axis shows defect area. Based on customer requirements the different coloured bands represent progressively worse defects, the bottom right corner being the worst. The customer can decide each individual band specifications and ultimately what is considered as acceptable. The crosses as detailed are defects positioned on the matrix according to size and contrast.

5. Commercial Review

Inspection results are displayed as a defect map on the fabric as shown above, together with a picture of the fault. In the case study given for workwear this was used to prepare an optimised cut plan for the fabric.

Manufacturers can immediately see:

• The proportion of each roll meeting required grade

• Whether issues are isolated or widespread

• Exactly where to cut or downgrade

• Which defect types are most costly

This improves:

• Speed — no manual searching for faults

• Objectivity — no inconsistent judgement

• Profitability — reduced scrap and claims

• Transparency — clear roll maps for customers

Although the system’s grading is objective, a final, commercial decision can be applied. A defect that is unacceptable for one product may be acceptable for another. Production leaders particularly value this pragmatic step.

Additional Capabilities 

Beyond defect detection and grading, the system can also measure:

Real-time skew

Fabric density

Orientation (face/reverse and direction)

Edge-centre-edge shade variation

Z-axis measurements including:

Pressure roller damage

Brushing pile height

Carpet pile height

Web thickness

Embossing depth

General surface texture changes

These are optional add-ons that extend the value of installation.

Applications 

In essence the applications for the system are limited only by imagination and return on investment. The list of textile sectors supplied is comprehensive including automotive, apparel, home textiles, industrial, medical, architectural, and military. 

The system can be fitted to fabric forming machines such as looms and knitting machines (see left). An alarm or machine stop can be initiated where a defect is detected as determined by the user. Regardless, the perch report with fault map guides menders to the faults speeding up this process. It also gives confidence that where fault levels are non-existent or minimal, fabric can go straight to the finish or despatch process. Statistical analysis of faults given by the Shelton Vision system leads to more confident and focussed quality improvement programmes.

Patterned fabric inspection is a recent addition to the functionality of the system and once again can be fitted to various points of the production process according to the user’s needs.

Above we see the cameras installed to a printing machine. The basic principle of the system is to have an image of the perfect print stored. The fabric print is then layered on the perfect print and one image subtracted from the other. Anything left is a fault.

The Shelton Vision system gives knowledge of fabric defects in a timely, accurate and automated way which allows, for example, fabric cutting plans to be optimised. When the entire supply chain builds confidence in the system it can lead to the situation where the garment value chain, for example, needs only one inspection. That would be a future worth having.

Summary & Conclusion

Automated fabric inspection has long promised improvements in quality, labour efficiency and customer confidence — but Shelton Vision demonstrated that the technology has now reached a maturity that makes widespread adoption commercially realistic. With high-speed detection, near perfect defect naming, objective grading, and clear roll-mapping, manufacturers can reduce waste, cut claims, and make faster and more profitable decisions on every metre produced.

As UK and global textile producers pivot toward higher-value markets and more efficient operations, automated vision systems such as WebSpector look set to become a standard part of modern finishing and quality assurance — marking a significant shift in how fabric quality is measured, managed and delivered.

For members of our textile societies, the question is how best can we harness the power of this system for the many small and medium enterprises that make up a significant amount of the UK industry. That is a question the Huddersfield Textile Society will look to help you answer in the future.

Questions were taken and audience thanked Mark Shelton and David Buxton of Shelton Vision for their lecture in the usual way.

This article is a report of the lecture at the Huddersfield Textile Society and is not intended as a recommendation of the Shelton Vision system. Other systems may be available. Graphics courtesy of Shelton Vision. No re-publication without consent.