Stabilizing quality in manufacturing with lean PLM and AI agents

In a previous blog article, I had explored how lean PLM supported by agentic AI can optimize manufacturers’ setup time and improve machine availability. By connecting PLM with real shop-floor execution, these businesses can continuously refine setup standards and unlock measurable improvements in Overall Equipment Effectiveness (OEE).

But availability is only one dimension of OEE.

The second major lever is quality.

Defects, rework and process deviations quietly erode productivity, consuming machine time, wasting material, and disrupting production flow. Many organizations try to solve quality challenges through engineering improvements alone, but the root causes are often deeply embedded in the daily realities of production.

During a consulting engagement with a spindle manufacturing company, this became very clear to me.

Here are some highlights, learnings and takeaways from this project.

A shop-floor lesson in quality

The manufacturer’s management team had a straightforward objective. They were tasked with increasing OEE by 10%. When we began our analysis, setup time quickly emerged as a major contributor to lost availability. That story had formed the basis of my previous blog.

But while spending time on the shop floor, another issue surfaced that was even more alarming. Nearly 50% of the produced parts were rejected.

The main causes were surface finish problems and misalignment during machining.

Initially, the assumption was that this was purely a machine problem. Perhaps the machine needed recalibration or the cutting parameters were incorrect. However, after speaking with the operators, a different picture emerged. One operator explained that once the machine started running, it usually ran continuously for hours. The real challenge was that operators were responsible for monitoring multiple machines simultaneously. Small deviations could start developing long before anyone noticed them.

A cutting tool might wear faster than expected.

A small vibration might appear in the spindle.

A fixture alignment could shift just enough to affect the surface finish.

None of these signals were dramatic enough to stop the machine immediately.

But they slowly accumulated. Parts continued to be produced until inspection finally detected the problem. By then, a significant number had already been impacted. That conversation helped us realize that quality problems were influenced by two categories of factors:

• Indirect factors related to workplace organization and operational discipline
• Direct factors related to the physical machining process itself.
  Both had to be addressed

Indirect quality influencers: The Role of 5S

Many quality issues originate not directly from machines but from the environment in which the process operates. This is where the 5S principles become extremely important.

Here’s what 5S stands for:

1. Sort – Remove unnecessary items.
2. Set in order – Arrange tools for easy access.
3. Shine – Maintain cleanliness and equipment condition.
4. Standardize – Define consistent ways of working.
5. Sustain – Maintain discipline over time.

In practice, however, sustaining a 5S discipline on a busy shop floor is not always easy.

Tools move from their defined positions. Spare fixtures accumulate near the machine. Gauges or measuring instruments are sometimes misplaced. Over time, these small deviations create friction in the production process. Operators spend time searching for tools. Incorrect tools may occasionally be used. Workstations become cluttered.

These indirect factors influence process stability, which eventually impacts product quality.

Today, modern digital technologies allow these indirect factors to be monitored much more effectively.

Vision systems can verify tool placement. Sensors can track workstation inventory levels. Digital checklists can ensure the correct fixtures and tools are available before machining begins.

Instead of relying on occasional audits, the system can maintain a continuous awareness of workplace conditions.

This is where lean PLM begins to extend its influence from engineering documentation into daily operational discipline.

Direct quality influencers: Monitoring the machining process

While indirect factors influence the environment, direct factors influence the physics of machining itself. During our spindle manufacturing investigation, two parameters repeatedly surfaced as early indicators of quality problems: Material Removal Rate (MRR) and excess vibration in the spindle system.

The MRR determines how aggressively material is being cut. If the rate changes due to tool wear or incorrect feed parameters, surface finish quality can deteriorate. Similarly, excessive vibration can indicate tool wear, imbalance, fixture misalignment, or early machine instability. These signals often appear long before the final quality inspection detects defects.

The above-mentioned parameters can be monitored continuously using modern edge systems that provide manufacturers with the following to make informed decisions:

• Machine controller data
• Vibration sensors
• Spindle monitoring systems
• Edge analytics

Detecting deviations early allows operators to intervene before large batches of defective parts are produced.

Agentic AI: Connecting direct and indirect insights

The real opportunity emerges when both direct and indirect factors are connected through a lean PLM architecture supported by agentic AI. In this architecture, PLM continues to act as the source of engineering knowledge, containing process parameters, tooling information, and work instructions. Shop-floor systems such as MES and machine controllers capture real execution data. Edge systems process signals from machines and sensors.

On top of this environment sits a layer of specialized AI agents that support execution and continuously learn from operational behavior. The architecture follows the same principles outlined in the agent-based workflow used to capture setup activities and improvements in lean PLM environments.

In a quality-focused scenario, these agents perform several key roles. Here’s what they do:

• A Workplace Monitoring Agent observes workstation organization, tool placement, and adherence to 5S principles.
• A Process Monitoring Agent tracks machining parameters such as material removal rate, vibration patterns, and machine signals.
• An Operator Guidance Agent provides contextual instructions and alerts when deviations are detected.
• A Root Cause Learning Agent analyzes patterns across production runs and identifies recurring causes of quality variation.
• A PLM Feedback Agent feeds validated improvements back into PLM to update process standards and work instructions.

Through this loop, PLM no longer acts only as a repository of engineering information. It becomes part of a continuous learning system connecting engineering intent with real shop-floor execution.

From availability to quality: Strengthening OEE

In the first article, lean PLM helped improve availability by reducing setup time. In this one, the focus shifts to quality stabilization.

By addressing both indirect factors such as 5S discipline and direct factors such as machining conditions, manufacturers can detect deviations earlier and prevent defects more effectively. Defect rates decrease. Rework is reduced. Production runs become more stable. Lean PLM begins to influence multiple dimensions of OEE simultaneously.

PLM that learns from production

The real transformation is not simply the introduction of AI into manufacturing. It is enabling PLM to learn continuously from real production environments.

Agentic AI closes the gap between engineering intent and operational reality by bringing knowledge, guidance, and insights directly into the flow of work.

Instead of relying solely on documentation and periodic improvement cycles, manufacturers can build systems where every production run contributes to stronger standards and better outcomes.

Lean PLM succeeds when planning and execution finally speak the same language.

>> Accelerate digital transformation for industrial value chains with Atos’s PLM services. Learn more today.

>> If you are ready to stabilize quality and boost OEE with Lean PLM, powered by Agentic AI, let’s discuss your quality challenges, 5S digitalization, and agentic AI use cases. Connect with me for your next steps.

Posted: 27/03/26