QU4LITY | Digital Reality in Zero Defect Manufacturing

We are building a connected environment through the industrial furnace smartization, but also implementing an IT solution that enables data gathering and transferring on real-time to GHI server, where then the data analysis is performed.

We are developing Sinapro IIoT MES/MOM cloud solution (part of the Kolektor Digital Platform) as the cornerstone of the MOM system which enables real-time collecting, evaluating, validating, filtering, checking, and storing of production data. The captured production data can be processed in real-time for the purpose of obtaining various production information, which enables immediate action. MOM function for production analyses with depth learning technology of AI gives users additional and high-quality information’s for fast decisions to achieve zero-defect goals in production. 

Operators are connected to the smart funcions the receive valuable information related to components condition that allow them to take decisions related to machine and process. Maintenance technitians and specialized engineers from the machine tool provider can also be included in the process.

Operators are more involved in the whole process, but mainly receives valuable information of the furnace operation through the real-time monitoring interface of the platform. Also daily reports provide them valuable information. 

The Kolektor Digital Platform enables human involvement on various levels. Human operators can monitor, view and inspect created datasets. During the process of model training, the operator can monitor the current state and detailed information of the training process. The Kolektor Digital Platform opens a channel between a data scientist and a decision-making individual in the production line. It is desirable to have multiple people, each assigned to a specific task. The whole process could be split into subtasks - acquiring images on the production line, human expert labeling the images (classification, anomaly,..), data scientist training the model on the new dataset and at the end evaluation of the model and pushing the new (improved) model to the production line.

The RiaStone Qu4lity Pilot enables human involvement at various levels and production stages.

Human operators inspect, monitor, and view all created datasets. During the process of model training, the operator guides the ML algorithm through the image inspection training process.

This is performed after the image acquisition in the production line, through joint algorithm/human expert labeling of the images (new defect classification, conformant/non-conformant product), this allowing for the training of the algorithm in the existing defects dataset, and at evolving the algorithm and pushing the new (improved) model to the production line AQL platform

Analyses Stations are equiped with Semantik Search Technology in order to generate consistent Data between standard failure catalogues and Teammember entries into Analyses Client. 

FAGOR’s monitoring platform, FA-Link machine platform, is being extended with tools focused on ZDM in this pilot and in the related packages of QU4LITY.

FA-Link is composed by two systems, the former that is executed in the manufacturing plant (on premise) and the latter that is executed in the cloud. The data captured in plant via on premise (local view), is uploaded and aggregated in the cloud part (global view).

The Beyond Platform also gathers data from other assets, not only from the furnace, so they can potentially optimised as well. On the other hand, there are aspects regarding the industrial furnace that mainly affects to the whole factory level, as for example, a reduction on the energy consumption or a reduction on the defective manufacturing that can be achieved thanks to the data analysis provided through this tools.

Analysing the test results the production process can be adapted and optimized.

The on-premise system is responsible of capturing the data from the different sensors and upload such information to the cloud. This task is performed by a software called FAGOR-DAS. Through FAGOR-DAS, data published by the sensors via PLCs using industrial protocols, such as OPC-UA, are sampled.  After this data is gathered, it is analysed and compacted locally (Edge computing). Such data can be visualized via a tool called Visual Stamp (local view of the manufacturing line). The same data is prepared to be sent to the cloud infrastructure.

The Beyond platform implemented provides valuable information with daily, monthly reports and deeper analysis in concrete operations that allows technicians to optimize the machine (industrial furnace) operation.

The Kolektor Digital Platform enables us to automatically collect the data from the shop floor. The Sinapro IIoT enables the connectivity of the pilot production line machines and related IoT devices for real-time production data acquisition and monitoring. The acquired data is afterward used in the off-line machine learning pipelines to produce machine vision predictive models to detect visual injection moulding defects. A pipeline for deploying such off-line machine learning to a HPC cluster is being developed at JSI within the scope of the Kolektor Pilot.

AI vision algorithm developed by TNO (WP3) seems to filter bad rated parts compared to installed algorithm. Advantage can be when product print is changing to catch-up development speed in traditional algorithm development. Test-case currently in progress.

Analyzing the test and process data, specific machine parameters can be adapted and optimized.

A correlation is realized within the production line between the overall process parameters and the product characteristics which are monitored at the end of the line in specific control modules.
With the help of INTRASOFT algorithm, several optimization are suggested for process parameters in order to optimize the final control workstation and to diminuate scraps and rework parts.

In addition to that, we can use CEA non-intrusive assets aquisition system to localize machine-oriented rootcauses (process deviation due to mechanical issue in the workstation for instance). This could lead to quickly identify a rootcause and to implement corrective actions effectively.
The non-intrusive notion comes from the fact that the asset monitoring (which can be a vibration from an accelerometer, a current, a temperature, ...) does not require any heavy integration, even the link with PLC is simplified with a standard exchange table for record triggers and the part informations.

FAGOR’s monitoring platform, FA-Link machine platform, is being extended with tools focused on ZDM in this pilot and in the related packages of QU4LITY.

There are two IoT platforms included in 2 manufacturaing lines for Automotive and Railway sector for MONDRAGON pilot. IoT platforms monitor 3 grinding machines from railway sector as well as press machine, stacker, Owen and Transfer from Automiotive customer. 

The variables monitored allow machine tool buider to know that the process exectuion is under treshold defined as well as enabling predictions about possible faillures. As a consequence there will be an increase of the quality production and the optimisation of the production. 

The interoperability layer between two IoT Platfroms has been achieved considering OPC-UA and AAS.

Collaboration with international partners such as ATLANTIS, VTT and FHG has allowed us to include IA algorithms, specific data analitics and data sharing connectors.In spte of the fact that IoT platfrom and interopoerability has been oriented to real time optimisation the analitycs and the IA algorithms have been carryed out off-line

With the quality management platform, optimization of production is enabled.

The optimization of quality process decision is taking place thank to a holistic view of the factors that influence the perception of the quality from the consumer prespective. The platform using a MPFQ driven data model is enabling a faster, more reliable and flexible visualization system and analytical approach.

Process Data from Manufacturing Data Lake is analysed online for suspicious outliers. A Webinterface enables online intervention by staff.

MONDRAGON pilot is being developed considering 2 IoT platfrom and interoperability layer developed by MGEP together with OPC-UA and AAS. Real time process optimisation enables Autonomous quality outcomes and Zero Defect Manufacturing for Automotve (Fagor Arrasate )and railway (Danobat) manufacturing lines. The FA-LINK platfrom monitored industrial assets for Fagor Arrasate and SAVVY IoT platfrom for DANOBAT.

The introduction of IA algorithms by ATLANTIS and VTT are developed offline achieving high top optimisation production. On the other hand, the approach of Machine Learning approach should be further developed. The interaction of the operators, maintenance workers and R&D staff are stil crucial for Top high level Autonomous Manufacturing process Optmisation

Part of the improved decision process enabled by the holistic platrom can be close looped into machine control parametes, allowing an autonomous quality management at factory level

1. Augmented Reality is improving supporting processes Change over,  Maintenance and Training. Partner PACE will apply their AR technology to avoid utilization of human resource in Maintenance documents handling. Instead Technologies like smart glasses and Holo lens will be applied. Virtual assistants will guide Maintenances staff through maintenance and repair processes instead. Same is targeted for Training.

2. Visualisation of machine and process data in realtime will enable immediate intervention in case of abnormal behaviour.

The acquired data is used in on-line prediction of defects. The predicted defects are used to adapt the visual quality inspection with an in-hand camera with a robot. The robot is guided to and between predetermined viewpoints associated with the predicted defects. The robot motion is generated autonomously on-line.

Thanks to this new approach with modular adaptable signal processing system and a strong interaction with data space and simulation tools trough the platform, will be possible to detect anomaly and have anequipment condition reporting , reduce reject rate by application of data-driven process model that has been derived by AI algorithms, increase OEE by recommending process adjustments to the operator or directly change the parameters in real time, so to reduce also the operator costs.

The data acquired through computer vision, is processed through machine learning algorithms and compared to an existing database of ~10000+ images already noted by human operators

Inspection results are fed into the synesis-consortium machine control platform that decides necessary changes to the machine parameters driving production in both Business Processes (1&2)

At a certain point of integration, the correlation between overall process parameters and the output product characteristics could be realized in real time in order to adjust, without any Human action, the different parameters along the production line.

This is the ideal towards which we wish to achieve with our future production lines.

Through the developments considered in this business case, the machine tool supplier obtains new digitally enabled solutions for the provision of smart services supported on digital platforms. 

By the time this sharing data solution has been devise just for the use case, but also a business model in here can emerge as GHI is getting more restrictions with customers that do not want to give the governance of their process data.

Data are valorized thanks to a noevl data model based on MPFQ wich is correlating in a function based structured way components parmeters to process parameters to quality performances

Data architecture is based on a Industrial Ontology derived from MPFQ model

the use o a standard ontology is the basic mechanism to provide a semantic meaning to all the data generated af shopfloor level and enable a urther high degree o correlation with all the other company genarated data.

improve false positive rate by 20%.  Measured as false positives rate, actual value is considered confidential.

Quality- Fall Off Rate, from 95% (as is) to 98,5% (to be)

Improve OEE (A) from 80%  (as is) to 87% (to be)

Improve OEE (A) from 80%  (as is) to 87% (to be)

Deviation on cycle-time, from 98% (as is) to 99% (to be)

Quality- Fall Off Rate, from 95% (as is) to 98,5% (to be)

Through innovative algorithms and statistical methods, possible data sources for predictive quality control can be identified and evaluated. Moreover, by cooperation of all project partners, the realization of data access and acquisition along the whole process chain can be realized. With a focus on algorithms and methodology, a use case-specific algorithm is going to be implemented and validated to maintain high prediction accuracy.

Data availability is a challenge: Limited access to measurement data (due to limited access to third-party systems)

There seems to be relationship to predict torque with use of in-line data. Needs to be more explored

By applying sophisticated algorithms and methods on the acquired data, systematic failure root cause detection supported by data analytics can be implemented. In addition, improved knowledge of machine states/maintenance requirements for neuralgic points can be implemented through the desired solution path within this pilot.

An AI vision algorithm developed by TNO (WP3) seems to filter bad rated parts compared to installed algorithm. Advantage can be when product print is changing to catch-up development speed in traditional algorithm development.

For this trial, the acquired test data will be analyzed regarding quality classification. In every test a part could pass or fail. Failed parts must be reworked, if possible, and brought back to the process. Sometimes parts are classified as failed even if they are good (false positive). This effect will be analyzed by machine learning algorithms and, if necessary, adopted in classification parameterisation. Additionally, the fact of 100% testing, means every panel is tested automatically, with bottleneck in out of the line test stations will be addressed in setting up failure prediction models for quality forecast. This will be supported by data analysis of pre reflow AOI (automated optical inspection). 

With all these data analysis and process optimization activities economical evaluation will be included to support decisions in-process and configuration changes. For the development of these applications, the main steps are data availability/access, data processing, and model development. The developed applications should be deployed on Edge devices.