Plastics Industry 4.0

2	Data Acquisition and Process Monitoring as Enabler for Industry 4.0

The most challenging tasks for an implementation of the self-optimization systems are on the one hand the identification of appropriate model-based optimization strategies and on the other hand the provision of required data from the process provided by the used sensors. [1]

Industry 4.0 claims to enable smartification of production systems through digitization and an enhanced introduction of information technology to traditional manufacturing processes. As the possibilities in production automation emerge, data becomes a valuable resource of the digital era. Therefore, as stated by Klocke et al., not only developing new ways to analyze data but also provision of the right data at the right time reflect major challenges of today’s research activities. However, advances in sensor technology and utilization of sensor data for process control have been in focus for a long time and can be expected to be the basis for current developments in process automation and digitization.

The origin of the data and the purpose of its acquisition determine whether it serves as valuable information to model e. g. an injection molding process in order to optimize the cycle time or to just monitor a continuous production with calculated key performance indicators. On the one hand, quality data like dimensions or part weight can be gathered to get information about the final part quality. On the other hand, machine data describes the dynamic machine behavior, such as screw movement or hydraulic pressure. The link between both, from machine data to the final part quality, is made through process data that describes the filling process inside the mold, e. g. by the cavity pressure.

In the Cluster of Excellence Integrative Production Technology for High-Wage Countries, for example, a model-based self-optimization algorithm has been developed that optimizes the operating point of an injection molding production based on the correlation of pressure, temperature, and specific volume (pvT-behavior) of the plastic melt (Figure 2.1). The ideal operating point therefore is determined by ensuring a constant specific volume during cooling in the holding pressure phase [2]. The model-based self-optimization utilizes the pvT-behavior as quality model to link cavity pressure and melt temperature as process data to the resulting quality data, a constant specific volume. For subsequent pvT-optimization, a voltage Ucontrol enables the control of the servo-inverter and the hydraulic valve to adjust the screw movement during the injection and holding pressure phase with respect to the demanded reference course of the cavity pressure pmold,ref. An appropriate process model therefore needs to correlate the machine parameters to the process data. However, disturbances like temperature variations of plastic melt and mold surface or viscosity fluctuations of the material affect the process behavior and impede a stable process. The model-based self-optimization compensates for these influences, for example by applying a sensor-actuator system as a model-predictive process control [2].

A quality model like the pvT-behavior links process and quality data. Subsequently, to optimize part quality, process data needs to be gathered and correlated to specific quality characteristics in injection molding. Whereas the filling process right inside the mold including the holding-pressure and cooling phase provides the most significant information to model the injection molding process with respect to a demanded part quality, the mold itself does not always offer the possibility to communicate this information. The mold therefore needs to have sensors to gather and transfer data about cavity pressure and temperature. For a single-cavity mold, for example, two combined piezoelectric pressure-temperature sensors, one near and one far from the gate, enable the monitoring of the flow path proceeding at each increment in time.

A process model like an Artificial Neural Network (ANN) or a simple model based on fluid mechanics with respect to the rheological behavior of the plastic melt correlates machine data and process data. Everything that happens inside the mold and that is measured as physical parameters describing the melt front velocity or a temperature distribution over the flow path is the result of mechanical movements that initiate a volume flow. Therefore, data about screw movements, screw torque, or the hydraulic pressure has to be gathered with respect to the resulting process data. Enhancing production automation and using digitization for realizing fully automated self-optimization algorithms as shown in Figure 2.1 requires a comprehensive view on the production system and a profound understanding of it as only machine parameters can be controlled directly. However, even with the possibility to calculate the control signal Ucontrol to realize a cavity pressure pmold that equals the cavity pressure reference course pmold,ref as described above, environmental fluctuations cannot be compensated completely and will always affect the production process.

Data from production management and logistics open up another different perspective. Considering more than one machine or production cell raises complexity to a completely new level as not only interactions between machine, process, and quality data have to be respected. In a Complex Value Chain, often several production cells interact with each other and human operators to fulfill a production order in a dynamic and efficient manner. Moreover, collaboration between different domains working on the same product or production order impedes optimization of production structures along the stages in order processing with respect to resource efficiency, profitability, or productivity.

To track the activities in a Complex Value Chain, the production order often is the only common element in a cross-domain and multi-process perspective. Information systems like an Enterprise Resource Planning (ERP) or a Manufacturing Execution System (MES) act as central data and information handling systems. Therefore, data gets centralized so that several domains get access with respect to their individual viewpoint such as in accounting, order processing, or material supply. The production order thereby acts as the one element to link every piece of information from production, assembly, or quality assurance. However, due to the necessity of making data and information accessible by a central information system, a consistent and persistent connectivity to field devices becomes a fundamental requirement in a production environment.

Standardized interfaces are expected to serve as a solution to realize a consistent interconnection between several devices and information systems. Especially due to various meta-standards in automation across different industries and upcoming new efficient communication protocols, choosing the right infrastructure becomes an increasingly complex task. Thereby, setting up the right infrastructure for individual Industry 4.0 applications is the source of great efforts and many discussions in the plastics processing industry.

Conclusively, data acquisition presents itself as an important and encompassing but necessary task when it comes to digitization of production processes. Therefore, it is not only the question of where to gather data (part, process, or machine) that is in focus, but also which technology is used as communication interface or in which information system it should be made accessible. Engineers as well as basic operators need to comprehensively understand the process and its physical boundary conditions to utilize accessible sensor technology for measuring the desired data and getting purposeful information about how to optimize and control the process. Data acquisition is furthermore a primary precondition of all data-driven applications and cybernetic process models. However, production technology tends to be so complex in structuring its assets and modeling, for example, the physical background of casting processes, that data-driven models alone will not provide sufficient solutions. The former Cluster of Excellence Integrated Production Technology for High-Wage Countries at RWTH Aachen University concludes that in complex, socio-technical production systems a combined approach of cybernetic models of data scientists and deterministic models of engineers enable an integrative comprehension (Figure 2.2). Therefore, cybernetic models support handling complexity and uncovering unknown phenomena and structures, whereas deterministic models reduce the complexity and identify subsystems that can be described by physical laws or physically motivated models [3].

2.1	The Necessity of Data Acquisition

Data acquisition is important to produce high-quality plastic parts. The use of data enables an improvement in the following areas:

       Control of process stability enables a fast reaction to process disturbances

       Optimization of the process: analyzing process data...