Zero-defect manufacturing is the culmination of a dream long cherished by the industrial sector. Production downtime is a key metric of manufacturing environments, and various quality methods and processes introduced over the decades have tried to reduce error rates. Now, Industry 4.0 methods of connecting operational machinery with digital intelligence are starting to deliver zero-defect assembly lines, ushering in a raft of new opportunities.

STREAM0D, a European Union-funded project aimed at implementing smart manufacturing processes within the context of Industry 4.0, has been piloting zero-defect manufacturing (ZDM) with three manufacturers in the automotive sector: Fersa Bearings; Standard Profil, a manufacturer of sealing systems; and ZF, a mobility systems supplier.

Integrating real-time simulation into factory production lines is a key aspect of Industry 4.0. José Ramon Valdes, STREAM-0D project coordinator, answered my questions about the outcomes of the pilot and the fresh opportunities it brings.

What are the business benefits of ZDM, quantified in the pilot?

JRV: It’s possible to:

  • Monitor and adjust the production process in real time, through simulation-based control algorithms and data-based decision tools, so that the product performance indicators fit the target specifications with an accuracy of 95 per cent, achieving zero defects in the final product.
  • Increase production flexibility by decreasing the time needed for the adjustment of new designs by at least 30 per cent.
  • Boost the efficiency of the line by achieving at least a 10 per cent reduction in rejected units and material.
  • Reduce production costs by at least 15 per cent and increase production rates by at least 15 per cent.

What improvements will zero-defect manufacturing bring to manufacturing processes?

JRV: It will allow them to:

  • Adjust the manufacturing process in real time, introducing smart decisions based on the forecast of the models. Injecting real-time intelligence into assembly systems can reduce product variability, increase efficiency, save time and costs, and reduce the number of rejected units or discarded material.
  • Absorb the effect of component variability, and even reduce the tolerance requirements for suppliers, as it will be possible to adjust process variables during the production process itself. Manufacturers can also exploit the full potential of knowledge-based simulation models like computational fluid dynamics (CFD), not only in the design process but also in the assembly process.
  • Increase the flexibility of manufacturing processes by setting design targets online through the control software, customising batches of units with different features, and reducing downtime related to changing design specifications.
  • Generate and exploit data-driven models built upon the stored data from measurements and models that will be used for further improving the process, detecting fault patterns, managing alarms and keeping traceability.

To what extent are simulations augmenting rather than automating operators’ decisions?

JRV: The uses vary:

  • In some cases, the simulation model is used to calculate values of adjusting parameters that are then plugged directly into the line, automatically adjusting the process without the operator’s intervention.
  • In other cases, the models are used to calculate and show recommended values of some parameters. The operator can then monitor a process and decide whether to apply the recommendations or not.

What kind of products and processes are being simulated on production lines?

JRV: Some pilot examples are:

  • A simulation model of a brake booster that predicts key indicators of booster performance, such as the pedal pressure. The simulator indicates required function of parameters that are measured in the line — dimensions of some components, material parameters of other components.
  • The production process of an extruded rubber profile for car doors and windows. We simulate how the shape of the profile changes as it is being pulled along the line, through the curing ovens and other stations.
  • A simulation model of a bearing component that predicts the exact dimensions as a function of the temperature. We also have a data-driven model, which isn’t strictly a simulation model, because its physics is not simulated. It’s a mathematical model built with historical data compiled in the line. This predicts potential defects and failures based on machine learning algorithms.

What are the lessons learnt so far in implementing simulations in production?

JRV: There are several lessons:

  • To build data-driven models, it’s absolutely necessary to have correlated data — matching input-output data — for every unit. This correlation is not always available and is sometimes difficult to implement.
  • Some input parameters needed to feed the simulation models are not easy to measure. Alternative or indirect ways have to be figured out and implemented.
  • Simulation models do not always reproduce the reality with 100 per cent accuracy. All models have a certain inherent inaccuracy: In-line recalibration methods have to be implemented, or models have to be enriched with data to improve their accuracy.