Over past months, many in-house workflows and development processes have slowed down and also changed. Nevertheless, time has not stood still, also not in injection molding technology. The debate and development have been shaped by two themes: digitalization and the circular economy. Below are some examples of developments and projects that underpin this trend.

Innovative Injection Molding in Times of Pandemic

Spritzgießen

Kunststoffverarbeitung: Verfahren

Schaumspritzgießen

Künstliche Intelligenz für die Qualitätssicherung

Industrie 4.0

Industrie 4.0 für die Qualitätssicherung

Steuern, Regeln

Software

Mehrkomponenten-Spritzgießen

Folientechnik (IMD, IML, FIM, u.a.)

Digitale Transformation

Wirtschaft

Innovationen und neue Technologie

Spritzgießwerkzeuge

Kunststoffverarbeitung: Maschinen und Komponenten

Werkzeuge für die Kunststoffverarbeitung

Biokunststoffe

Kunststoff Werkstoffe

Spritzgieß-Sonderverfahren

Nachhaltigkeit

Umweltmanagement

Spritzgießsimulation

Simulation

Schnecken und Plastifiziersysteme

Spritzgießmaschinen

Industrie 4.0 für die Kunststoffverarbeitung

Machine Learning / Künstliche Intelligenz für die spanende Fertigung

Industrie 4.0 für die spanende Fertigung

Kunststoff-Recycling

Kunststoff-Aufbereitung

Lasermarkieren

Oberflächentechnik

Laserbearbeitungen

Inline- / Online-Messsysteme

Qualitätssicherung

Qualitätssicherung Kunststoff

Kreislaufwirtschaft

There Is More beyond the Horizon

In an online survey, representatives of the plastics industry formulated their expectations toward the Plastics Innovation Center 4.0 (PIC 4.0).

Part 5 of the Series: Practical Demonstrators for the Plastics Innovation Center 4.0

Messdatenerfassung und -verarbeitung

Prozessoptimierung

Prozessmanagement

Machine Learning

Datenanalyse

Automatisierungstechnik

Kunststoff-Rezyklat

Focus on the User Benefit

The latest trend in digitization  is the Artificial Intelligence of Things. It amounts to “marrying” Artificial Intelligence with the Internet of Things. In the following interview, Martin Dimmler and Martin Rugfelt discuss how companies can profit from AIoT and what plastics manufacturers can learn from its pioneers in other industries.

Artificial Intelligence of Things: Trends and Models for the Plastics Industry

Vernetzung / Cloudlösung für die Kunststoffverarbeitung

IoT-Plattformen für die Qualitätssicherung

Fernwartung / Ferndiagnose

Instandhaltung

Produkt- / anbieterspezifische Automatisierung

Automatisierungszellen

Energieeffizienz

Fertigungsautomation

IoT-Plattformen für die Kunststoffverarbeitung

“Intelligent Automation Brings Systems Closer to their Optimum”

Part 2 of the series on the Plastics Innovation Center 4.0 (PIC 4.0) at RWTH Aachen describes the procedure by which industrial use cases of a Smart Factory are identified and converted into real testbeds within the PIC 4.0.

PIC 4.0 Elaborates Holistic Solutions for Industrial Demands

Use Cases of Digitized Plastics Processing

A consistent flow of data and information is essential for digital solutions and meaningful analytics. The introduction of potential solutions is therefore often associated with high costs for data preparation.

Standardization Guarantees High Data Quality for Future Applications

To Get the Information Flow in a Row

Once again, the K 2019 in Düsseldorf, Germany, lived up to its claim to be the leading trade show for plastics technology. At least in the field of injection molding, there were exciting global innovations to see – if you looked closely.

Injection Molding Trends at K 2019

Smart Factory

Mikrospritzgießen

Spritzgießmaschinen, allgemein

Fließpressen

Handling, Handhabungstechnik für die spanende Fertigung

Handling, Handhabungstechnik

Elastomer-Spritzgießen

Linearroboter für die Kunststoffverarbeitung

Industrieroboter

Montagetechnik

Montage-Spritzgießen

Knickarmroboter

Co-Injektion (Barriere-/Sandwich-Spritzgießen)

Hybrid-Spritzgießen (Metall / Kunststoff)

Ressourcenmanagement

Produktion

Technical Excellence thanks  to Digital Assistance

The Reversecube two-part cube mold is used to produce 60 million multicomponent parts per year on one production line and in the same process. The cycle times must be kept very short while at the same time maintaining stable processes and a high part quality. Sensor-based monitoring and automatic process control is therefore deployed in the form of Fillcontrol from Priamus. It consists of hot-runner balancing, viscosity and compression control as well as automatic holding pressure changeover.

Automating Injection Molding with Process Control – with the Example of a Cube Mold

Werkzeuginnendrucksensoren

Messtechnik

Sensoren / Messwertaufnehmer

Werkzeugtemperatursensoren

Quality Assurance for Multiple Cavities

After a year’s forced break of Fakuma, Wittmann Battenfeld is announcing the new servohydraulic SmartPlus machine as the highlight of this year’s trade show. It is claimed to be characterized by high economic and energy efficiency and repeat accuracy.

Messe Fakuma

kunststoffe-international.com

Spritzprägen

Injection Molding Innovations In-House, Live and in Color

The Circular Economy is not only the theme topic at the K 2019. It is also the “green” thread running through the fair exhibit by Engel, the provider of injection molding machines and integrated system solutions. Industry 4.0 is turning out to be a driving force for recycling. This is a most timely insight, since the branch is in for several economically cool years.

Engel Supports Strong Involvement by the Plastics Industry in a Circular Economy

Doriat, Clemens, Dr.

Verbundwerkstoffe / Werkstoffverbunde

Organobleche (gewebeverstärkte Thermoplast-Halbzeuge)

Halbzeuge und Zwischenprodukte

Kunststoff-Halbzeuge

A Second Life for Plastics

During this year’s pandemic, maintaining supply capability has become a key necessity among processors. Closer involvement of suppliers is only logical in these circumstances. In toolmaking, new challenges are emerging for services related to molds.

The Advantages of a Network for Global Qualification for Injection Molding

Erstbemusterung, Erstmusterprüfbericht

Werkzeug- und Formenbau

Dienstleistungen

Werkzeug-, Formen- und Modellbau

Qualitätsmanagement

Werkzeugbau

Fast and Effective Start of Production

The toolmaking industry of the future is facing significant changes in its range of services and its value creation processes. It is difficult to predict exactly how the collaboration of German tool manufacturers with series production will be structured in the future. However, it can be assumed that digital networking and sustainability will play a key role.

What the Two Trend Topics Mean for Collaboration between Toolmaking and Series Production

Vorbeugende Instandhaltung für die Qualitätssicherung

Toolmaking of the Future –  Digitalized and Sustainable

Because it has not yet been possible to exploit the full potential of digitization in plastics processing, the Institute for Plastics Processing at RWTH Aachen University is building a new application-oriented research infrastructure along the lines of a Smart Factory. Kunststoffe is accompanying the building of the Plastics Innovation Center 4.0 with a series of publications. Part 1 describes the concept and the initial demands on the physical infrastructure.

The Plastics Innovation Center 4.0 Is under Construction at the Campus Melaten in Aachen, Germany

Central Hub for Digitization  in Plastics Processing

Munich-based mechanical engineering group KraussMaffei is repositioning itself and uniting all divisions under a single brand. At a time of accelerated technological change, the company is increasing its efforts to develop new products for digital solutions and new business models – and that includes plastics recycling.

KraussMaffei Consolidates Three Divisions and the Previous Group Brands under a Single Brand

Extrusion

Compoundieren

Moving with the Times

M-Hub is the first mobile App that helps entrepreneurs to promote their business and find all players in the plastic industry with the tap of a finger.

The Plastic Industry in Your Hand

The cluster of excellence “Internet of Production” at the RWTH Aachen University conducts research on structures to realize a “Google of Production”. The Plastics Innovation Center 4.0 serves as demonstration platform to present real-world solutions to the industry already today.

IKV Offers an Application-Oriented Demonstration Platform for Excellent Research

The PIC 4.0 as Part of the Internet  of Production

Inline Quality Monitoring and  Data-Based Quality Prediction

Successfully Managing Digital Transformation with Automation and Artificial Intelligence

Maintaining competitiveness while producing in Germany requires not only cost-efficient production, but also high-quality products. Vorwerk Elektrowerke is accepting this challenge. In collaboration with the Institute of Product Engineering at the University of Duisburg-Essen, Germany, the company is testing a 100 % inline quality monitoring system as well as a quality prediction using machine learning.

The Vorwerk Elektrowerke GmbH & Co. KG manufactures high-quality household appliances that are known worldwide under brand names like “Thermomix” and “Kobold”. Since these are primarily located in the upper price segment, customers have correspondingly high demands on product quality. These are reflected in the quality requirements that apply to the molded parts used. These are produced in three shifts at the Wuppertal plant in Germany, the largest production site of the Vorwerk Group, mainly by injection molding.

Production at the high-wage Germany generally requires continuous innovations in the field of process technology in order for the company to remain competitive in the long term, both in terms of quality and cost aspects. Within this context, Vorwerk has decided to work together with the Institute for Product Engineering (IPE) at the University of Duisburg-Essen to investigate the added value of optical 100 % inline quality assessment and direct prediction of molded part quality based on process data for injection molding production.

The first step was to select a suitable molded part. The choice fell on the filter cover of the Kobold VK 200 vacuum cleaner (Title figure). This is made of acrylonitrile-butadiene-styrene (ABS) and has a sealing lip made of a thermoplastic elastomer (TPE). The filter cover (Fig. 1) is produced on a hydraulic two-component injection molding machine (type: 500– 2000–180-CX Z; manufacturer: KraussMaffei Technologies GmbH, Munich, Germany).

In addition to the assembly capability and an optically perfect surface of the molded part, the sealing effect of the sealing lip is of great importance. Otherwise, the function of the vacuum cleaner would be affected, and the customer would be bothered by unwanted background noise when air flows through the sealing area. Two things are necessary to ensure the sealing effect:

<ul><li>The complete forming of the sealing lip and</li><li>the accurate, circumferential position of the TPE contour.</li></ul>

The latter is directly related to the molded part contour of the hard component, which is represented by two width dimensions and should be the focus of this investigation.

Inline Assessment of the  Molding Contour

A digital camera system (type: CV-X; manufacturer: Keyence Deutschland GmbH, Neu-Isenburg, Germany) was used for the trial (Fig. 2). As no color information is required for recording the molding contour, a monochrome sensor came to use. In order to be able to map the variation in width dimensions sufficiently accurately, the highest-resolution sensor in this model series was chosen, with 21 megapixels. The entire camera system was integrated into the existing process both in terms of hardware and software by Mabri.Vision GmbH, Aachen, Germany.

The molded parts are photographed immediately after removal, while they are still in the robot’s grip. Therefore, the handling robot sends a trigger signal to the camera control. Afterwards, the molded part is fed to the assembly line via an automatic conveyor system. Over an investigation period of around four weeks, more than 22,000 molded parts were produced, and their quality parameters recorded as described above.

Overall, the system proved to be very robust. It is designed in such a way that it automatically starts when the robot is switched on and records the quality of all parts produced from then on. The molded part dimensions extracted from the recorded images were defined once, then automatically recorded and provided with a time stamp. For trouble-free further processing, it is therefore important to synchronize the times of the injection molding machine and the camera system. This can be ensured in the case of permanent use by connecting the system to a time server through connection to the company network.

Within the scope of optical quality monitoring, two width dimensions are recorded for fundamental testing. These are determined as the maximum perpendicular distance from the middle partition of the molded part to the outer contour and are referred to in the following as dimension 1 and dimension 2. The total width of the molded part is the sum of the two individual dimensions. By recording the individual dimensions, statements can be made about possible asymmetrical warpage of the part in addition to the overall width, which is equally relevant to quality.

In addition to the inline quality monitoring described above, the project team investigated whether the quality parameters under consideration can also be predicted indirectly – i. e. without measurement – from the process data stored in the actual data log of the injection molding machine. For this purpose, methods from the field of machine learning are used, which is a sub-area of artificial intelligence. Although this topic has already been investigated in the past [1–4], available applications [5] have so far hardly been able to establish themselves in industry.

Machine Learning for the Prediction  of Quality Features

Based on many years of experience in the analysis of injection molding process data, the IPE researches the various aspects relevant for molded part quality prediction. The aim is to create a holistic system that performs all necessary data processing steps automatically [6] and therefore enables the user to handle the system easily (Fig. 3). The procedure comprises in detail:

<ul><li> Data pre-processing,</li><li> feature construction and selection,</li><li>modeling and hyperparameter optimization,</li><li>model selection and quality prediction with the best model, as well as</li><li>model adaption in case of changed process conditions.</li></ul>

The data preparation (standardization, etc.) required for proper processing is followed by the construction and selection of the process features used as input for the quality prediction, the first central step for a successful prediction [7]. Own investigations [8] have shown that a filter-based approach with sequential forward selection as search strategy and correlation-based feature selection [9] provides the best results in injection molding.

Seven “state of the art” learning algorithms from the field of supervised learning are used for the actual modelling. In addition to well-known algorithms such as artificial neural networks, e. g. Support vector machines and Gaussian process regression are included. These are trained in parallel as part of the overall system and compete against each other – only the model that achieves the best performance on the validation data is finally used for the prediction.

To ensure that the individual learning algorithms reach their maximum performance, the respective hyperparameters that control the model structure and learning process are automatically optimized during modeling. Through production, the relationships between process and quality parameters can change, which reduces the quality of the prediction. This can be recognized by methods which can detect so-called concept drift, which also were implemented. The IPE was able to design the prediction system in such a way that operation on a desktop PC of the current generation is possible with reasonable computing effort.

To evaluate the quality prediction, a design of experiments (DOE) covering 75 injection molding cycles was carried out in addition to the approximately 22,000 cycles from series production. The following machine setting parameters, selected based on experience, were varied within the scope of the DOE:

<ul><li>Injection speed of the hard component (ABS),</li><li>holding pressure level of the hard component,</li><li> cooling water temperature,</li><li>cooling time.</li></ul>

Because of the lower number of cycles in the DOE, it was possible not only to assess the quality features with the inline camera, but to manually measure with a gauge specially manufactured for this molded part, which is also used in normal serial production. This allows a comparison of how well the quality data generated with both measuring methods are suitable for modelling. It must be considered that the shrinkage of the molded parts is not yet completed at the time of recording with the camera system, so that there is a deviation from the final dimensions.

It should be noted that modelling is generally also possible based on historical process and quality data. The former is continuously recorded in any case by the injection molding machine. However, the condition is, that the quality data is assigned to the exact process cycle.

The coefficient of determination is used to evaluate the models. This can achieve values between 0 % and 100 % and indicates the degree of overall variance of the quality feature described by the model. The results show that a quality prediction with coefficients of determination of up to 98.8  % is generally very well possible (Fig. 4).

In the DOE, the model performance based on the camera data with coefficients of determination between 95.7 % for dimension 1 and 98.8 % for the total width are slightly above the model performance on the basis of the gauge data with 87.8 % for dimension 1 to 93.6 % for the total width. The coefficients of determination for the prediction of dimension 2 lie in between. However, all these model performances can be classified as satisfactory.

In serial production, however, the coefficients of determination of the two partial widths (dimension 1 and dimension 2) are significantly below the coefficient of determination of the overall width. The latter has a very high model performance of 90.0 %. As a further analysis showed, the low coefficients of determination of the part’s partial widths (12.6 % and 18.1 %) are due to an insufficient accuracy of the camera system when recording the position of the middle rib, used as a reference. This occurs when the alignment of the part on the gripper is subject to fluctuations around the vertical axis and could be remedied by adjusting the measuring system.

Since the errors of the partial widths cancel each other out exactly, the overall width is unaffected. In the case of the DOE, this problem has hardly any negative effect, due to the variation of the machine settings, the part’s real variation outweighs the measurement uncertainty. Overall, it can be stated that modelling is also possible on the basis of data from series production, as long as there is sufficient measurement accuracy.

All in all, the conditions for the future use of the technologies investigated are very good, says plant manager Dr.-Ing. Markus Koch (Fig. 5): “Competitiveness in Germany as a high-wage location can be achieved if the autonomy times of manufacturing processes are high and therefore employee deployment is low. If this succeeds with low reject rates, as is the case with permanent quality monitoring, the duty is fulfilled. The freestyle is achieved when process adjustments can be made depending on quality deviations without the need for manual intervention.”

With the investigation of the inline camera system and the model-based molded part quality prediction, Vorwerk Elektrowerke decided to actively shape the digital transformation and thus secure its competitiveness in the long term. The camera system proved to be easy to operate and very robust over the test period. The testing of the quality prognosis system was also successful. The high model performance shows that it is possible to make a good prediction of the quality characteristics of molded parts. However, this requires a high accuracy of the quality data used in the learning phase.

1 Häussler, J.: Eine Qualitätssicherungsstrategie für die Kunststoffverarbeitung auf der Basis künstlicher Neuronaler Netzwerke. PhD-Dissertation, Universität-Gesamthochschule Essen 1994

2 Vaculik, R.: Regelung der Formteilqualität beim Spritzgießen auf Basis statistischer Prozessmodelle. PhD-Dissertation, RWTH Aachen 1996

3 Al-Haj Mustafa, M.: Modellbasierte Ansätze zur Qualitätsregelung beim Kunststoffspritzgießen. PhD-Dissertation, Universität-Gesamthochschule Essen 2000

4 Schnerr, O.: Automatisierung der Online-Qualitätsüberwachung beim Kunststoffspritzgießen. PhD-Dissertation, RWTH Aachen, 2000

5 Liedl, P.; Haag, G.; Müller, H. et al.: Spitzenqualität mit kurzen Zyklen. Kunststoffe 100 (2010) 2, pp. 38–40

6 Schulze Struchtrup, A.; Kvaktun, D.; Schiffers R.: A Holistic Approach to Part Quality Prediction in Injection Molding Based on Machine Learning. In: Hopmann, C.; Dahlmann, R. (Eds.): Advances in Polymer Processing 2020. Springer Vieweg, Berlin, Heidelberg 2020

7 Guyon, I.; Elisseef, A.: An Introduction to feature extraction. In: Guyon, I.; Gunn, S.; Nikravesh, M. et al. (Eds.): Feature extraction – foundations and applications. Springer Verlag, Berlin, Heidelberg 2006

Universität Duisburg-Essen Inst. f. Produkt Engineering 

https://www.uni-due.de/kkm/

Vorwerk Elektrowerke GmbH & Co. KG  

https://www.corporate.vorwerk.de

Ausgabenartikel

Bild 5. Continuous quality monitoring as a must, autonomous process control as freestyle – plant manager Dr.-Ing. Markus Koch is satisfied with the investigations

Customers who choose the Kobold VK 200 vacuum cleaner expect top quality. To ensure this, efforts for the introduction of Industry 4.0 measures are taken in the production plant in Wuppertal, Germany

16:9

Fig. 1. The contour of the molded part is crucial for the functionality of the sealing lip (TPE) of the filter cover

Bild 2. The CV-X camera system used has a monochrome sensor with a resolution of 21 megapixels

Bild 3. The automatic model generation relates to only little effort for the user: once the learning data has been provided, everything else runs automatically

Bild 4. The achieved model performances show that a prediction of the quality characteristics based on the process data is well possible. The requirements for this are precise quality data 

Name - Hauptartikel

The Vorwerk & Co. KG was founded in Wuppertal, Germany, in 1883 and has developed from a carpet factory into a broadly-based international group of companies in the course of its more than 130-year history. Vorwerk’s core business comprises the development, production and sale of high-quality household products. Vorwerk Elektrowerke is part of the family-owned company that employs a total of more than 12,300 permanent staff. As the largest production location of the group with its headquarters in Wuppertal, Vorwerk Elektrowerke manufactures high-quality household appliances such as the multifunctional Thermomix food processor or the Kobold vacuum cleaner.

<a href="http://corporate.vorwerk.com">corporate.vorwerk.com</a>

Alexander Schulze Struchtrup, M.Sc., is group leader injection molding since 2018 at the Institute of Product Engineering (IPE) at the University of Duisburg-Essen, Germany;

<a href="mailto:alexander.schulze-struchtrup@uni-due.de">alexander.schulze-struchtrup@uni-due.de</a>

Dipl.-Ing. Alexander Mahl is a project engineer in plastics technology at Vorwerk Elektrowerke GmbH & Co. KG since 2015.

Dimitri Kvaktun, M.Sc., is a research associate since 2019 at IPE.

Dipl.-Ing. Martin Thalemann is group leader plastics technology at Vorwerk since 2015.

Prof. Dr.-Ing. Reinhard Schiffers is holding the chair for Engineering Design and Plastics Machinery since 2017 at IPE.

The authors would like to thank Mabri.Vision GmbH, Aachen, Germany, and Keyence Deutschland GmbH, Neu-Isenburg, Germany, for the good collaboration during the installation and testing of the inline camera system.