In many factories, quality performance continues to be managed with partial visibility. Scrap rates increase, rework accumulates, operators grow tired, but the real causes remain difficult to objectify.

A rejected part can actually be compliant. Another, truly defective, can slip through the inspection. Between ever-higher customer requirements, industrial cadence, and human variability, maintaining a stable level of quality becomes a constant balancing act.

The problem is often deeper than a simple lack of detection. In reality on the ground, many manufacturers still work with incomplete quality specifications, orally transmitted criteria, or visual inspections reliant on the experience of the teams. The result: quality becomes variable from one operator to another, from one shift to another, or even from one hour to another.

Optimizing quality performance therefore does not consist solely of detecting more defects. It is primarily about stabilizing quality decisions, understanding drifts earlier, and reducing the cost of industrial non-quality before it becomes visible to the customer.

Why scrap and rework skyrocket in some factories

In many industrial environments, quality control still acts as a final filter rather than as a process management tool.

Operators sometimes inspect several parts per second under strong cadence pressure. In the automotive, plastics, or cosmetics packaging sectors, this reality mechanically generates:

• false rejects,

• undetected defects,

• variability in judgment,

• and hidden costs rarely measured precisely.

Visible scrap often represents only a fraction of the problem. Indirect costs are sometimes more significant:

• rework,

• line slowdowns,

• manual over-inspection,

• customer complaints,

• quality disputes,

• analysis time,

• loss of internal trust.

For some manufacturers, Spark, the AI-powered automated quality inspection solution developed by Scortex, has also highlighted drifts that were previously completely invisible: supplier defects, tooling drifts, or material instability.

For example, a cosmetics packaging manufacturer working with glossy surfaces noticed a high rate of false positives caused by dust and micro-reflections. At another luxury sector client, operators proactively rejected compliant parts to avoid customer complaints.

In this type of context, quality performance becomes difficult to maintain sustainably with manual inspection alone.

The real problem: quality is often subjective

One of the most common findings observed on the ground concerns the lack of truly actionable quality specifications.

Many factories have old, incomplete, or overly theoretical quality documents. Some criteria remain implicit:

• what is an acceptable Defect?

• at what size does a scratch become critical?

• is a slight color variation acceptable?

• should a micro-scratch close to a logo be rejected?

In premium cosmetics, for example, this subjectivity becomes even stronger. A simple lack of gloss on a lipstick can be judged critical because it gives the impression of a pre-used product.

Spark often acts as an eye-opener for these inconsistencies. When a system applies stable rules at a constant cadence, discrepancies in interpretation immediately become visible.

Automating quality control then forces teams to clarify:

• real tolerance thresholds,

• critical defects,

• acceptable anomalies,

• and the level of severity expected depending on the product.

This step is crucial for sustainably improving quality performance.

Reducing scrap without increasing escapes

One of the major industrial challenges is finding the right balance between: false positives and escapes.

A system that is too strict generates useless scrap.
A system that is too tolerant lets defects slip through.

In the automotive, luxury, or medical device production sectors, for example, manufacturers often favor an over-quality approach in order to limit any customer risk. Conversely, some high-speed lines seek a more productive balance.

The goal of automating quality control is not to replace operators or quality teams, but to reduce the variability of the first inspection level so that operators can focus on higher value-added tasks in the factory.  

Why data analysis completely changes quality performance

Sustainable quality improvement does not come solely from automatic sorting. It comes primarily from the data generated by the inspection.

Each inspection performed with Spark, an AI-powered automated quality inspection system, produces:

• an image,

• a heatmap (heat map to visualize anomalies),

• an OK/NOK result,

• a tracking of the scrap rate over time,

• a timestamped history,

• and information on the detection severity of anomalies.

This database then allows identifying:

• recurring defects,

• rejection peaks,

• process drifts,

• variations related to a supplier,

• or even critical areas of a part.

With one of our industrial clients manufacturing glass containers, Spark made it possible to correlate a defect peak with a furnace temperature issue. The setting was quickly adjusted before massive scrap production occurred.

In another case, a client manufacturing metal parts identified a material anomaly coming directly from the coil supplier.

Optimizing quality performance today requires more than a simple visual inspection. The most successful manufacturers are now looking to reduce scrap, stabilize quality decisions, understand drifts earlier, and capitalize on their inspection data.

AI-based systems enable precisely transforming inspection into a continuous improvement tool, while relieving teams of the most repetitive tasks. To go further, you can download the comparison Spark vs other quality control solutions, the Spark datasheet, or the Industry and AI guide.

FAQ

How to reduce the cost of non-quality?

The first lever consists of detecting drifts earlier. A stable and traceable inspection avoids the production of fully defective batches, massive rework, and customer complaints.

How to reduce scrap without slowing down production?

The main challenge lies in detecting drifts early enough to avoid full series of non-compliant parts. AI-automated inspection allows stabilizing quality decisions at a high cadence while limiting false rejects.

How to reduce rework in industrial production?

Rework often stems from defects detected too late or quality criteria that vary among teams. By standardizing control thresholds and leveraging inspection data, manufacturers can identify recurring causes and correct drifts faster.

What is the difference between classic machine vision and AI?

Classic vision applies fixed, pre-programmed rules. An AI like the one in Spark learns the normal variability of compliant parts and detects unusual discrepancies, even when they have never been seen before.

Here are other articles that might interest you:

·        Reducing hidden costs through automated visual inspection

·        How to structure a quality defect library to secure your decisions

·        Cosmetics and packaging quality control: making the inspection of demanding surfaces reliable thanks to AI