Synthesis and Analysis of Quality Control Methods for Intelligent Processing of Polymeric Materials

Global manufacturers of thermoplastic molded parts increasingly require 100% quality inspection levels that are difficult to achieve. While process complexity makes it difficult to attain the desired part properties during start-up, the stochastic nature of the process causes difficulty in maintaining part quality during production. This paper formally compares several alternative quality control methods that are currently utilized for processing of polymeric materials. To identify the technical issues associated with this goal, the injection molding process is described utilizing a control systems approach. Afterwards, four different methods of quality regulation are synthesized for injection molding: open loop quality control, statistical process control, trained parameter control, and on-line quality regression. For each strategy, the level of quality observability and controllability are determined against the dynamics of the manufacturing system. The results indicate that none of the quality regulation strategies have the underlying design architecture to deliver 100% quality assurance across a diverse set of application characteristics (quality requirements, material properties, mold geometries, and machine dynamics). As such, subsequent discussion focuses on defining the system requirements for achieving ‘intelligent’ processing of polymeric materials that are needed by industry. NOMENCLATURE a Acceleration of ram Amelt Cross-sectional area of barrel subjected to melt pressure Ahyd Cross-sectional area of injection cylinder submitted to hydraulic pressure C Volume of charge stroke as measured in total ram displacement Kp Process transform between ram position and melt pressure Kv Velocity compensation for ram velocity control m Mass of ram s Laplace transform Pmelt Melt pressure entering mold Q_dimension Quality of part dimensions: low indicates acceptable, high indicates defect Q_flash Existance of flash: low indicates acceptable, high indicates defect Q_short Existence of short molding: low indicates acceptable, high indicates defect u Desired ram velocity v Observed velocity of ram x Position of ram INTRODUCTION Injection molding of thermoplastics has emerged as a premier vehicle for delivering high quality, value added commercial products. Perhaps due to this success, continued global competitiveness has increased standards for product quality while requiring reduced product development time and unit cost. Despite advanced product design methods and new process technologies, it is becoming apparent that the injection molding process is neither flexible nor sufficiently robust to meet these industry requirements. The lack of robustness is sometimes evidenced by long product development cycles, excessive tooling costs, low process yields, and inferior product quality. As thermoplastic materials continue their thrust into advanced technical applications with multiple stringent requirements, the risks of proving out the injection molding process are becoming excessive. In fact, several industry managers have independently testified that “we are already starting to see the migration of customers to other manufacturing processes for time-critical applications.” Fundamentally, the difficulties associated with injection molding arise from the lack of simple and consistent relationships between the machine inputs, part geometry, material properties, and molded part quality. In product and tool design, numerical simulations have been developed to aid the design engineer. In tuning and regulation of injection molding, however, no method has had similar success in aiding the process engineer. Polymers exhibit extremely complex material properties – non-Newtonian, non-isothermal, thermoviscoelastic rheology together with highly temperature and pressure dependent thermal properties. During processing, the material undergoes temperature and pressure increases, significant shear deformation, followed by rapid decay of temperature and pressure in the mold cavity, which leads to solidification, and locking of residual stress, orientation, and other part properties that determine the molded part quality. While this process complexity makes it difficult to attain the desired part properties during start-up, process variability causes difficulty in maintaining part quality during production. The difficulty controlling this process’ phenomenological behavior is further complicated when coupled with diverse application requirements encountered in industry, i.e. every commercial application generates a unique set of process dynamics and quality issues. The objective of ongoing research is to fill this gap by developing methods for intelligent processing of polymeric materials. Desired Capabilities of Intelligent Process Control Definition I: Intelligent process control enables the manufacture of world-class quality product without expert knowledge of plastics by the design, tooling, or molding personnel. By this definition, intelligent process control (IPC) is both a product development strategy as well as a quality regulation strategy. The desired capabilities of IPC extend beyond quality regulation: • applications should be designed so as to be manufacturable to a process capability index of two, corresponding to Motorola 6σ quality levels (Alsup, 1993); • mold pre-production and process commissioning should be achievable without experts and require a minimum amount of time and tool modifications; • automatically estimate and assure quality of molded parts, or advise source of defects if not possible. These three goals form a set of capabilities that are widely sought after in industry. The first capability, application design, is largely the domain of computer aided engineering (CAE). The second bullet, mold commissioning, is the domain of expert systems and design of experiments. While not fully realized, these areas are well developed commercially and already add significant value to the product development process. The third capability, quality assurance and control, is insufficient for the characteristics of the molding process. The effectiveness of the system to assess the molded part quality is critical – the quality costs induced by poor quality prediction and control can quickly exceed the possible benefits of the quality control system (Morse, 1987). SYSTEM DESCRIPTION Various strategies for quality control will be discussed with the control system schematic shown in Figure 1. In control terminology, the primary two components of the system are the plant and the controller. The plant consists of the actuation components of the molding machine, the mold geometry, the material properties, and the quality dynamics of the molding application. The quality dynamics are determined by physical laws and thus can be designed but not controlled. As implied within the machine schematic, there can be multiple internal process controllers that regulate the machine actuators within a molding cycle. The Controller Quality Controller Quality Predictor Desired Quality Quality Dynamics Machine Actuators Process Controller Process Sensors Process Dynamics Observed Quality The Plant The Machine Noise Figure 1: General System Representation of the Injection Molding Process The second major component of the molding system is the controller, which compares the observed manufactured product properties to the desired part quality. As will be discussed in the next section, the controller can be human or machine, open or closed loop. In all cases, however, the controller attempts to bring about the desired part quality by setting the machine parameters or set points. The decision of when and how to change the inputs to the plant are based on some form of quality predictor, which can be simple or complex, empirical or analytical. The most significant input to the molding process is part quality, which is immediately converted by the controller (usually a process engineer) to a set of process parameters. These process inputs include machine parameters (such as temperature, velocity, and pressure profiles), material properties (rheological, thermal, and compressibility behaviors), and mold geometry (part topology, configuration, and parameters). An undesirable but second set of inputs to the molding process is noise, both intrinsic (natural variation about the process set points) and extrinsic (longer-term fluctuations of uncontrolled parameters). Noise can be observed but is generally expensive to convert to a controlled input. For example, relative humidity in the plant air might have an observable effect of the molding process and molded part quality. Moreover, relative humidity can be controlled albeit at a fairly significant cost. Continuing the discussion of Figure 1, all the molding process inputs are converted to process conditions that act on the polymer melt. Many of these process conditions (such as melt temperature, melt pressure, cavity pressure, etc.) are observable and related to the molded product quality. However, the exact relationships between the process conditions and the output quality are not always available. Rather, it is the internal process dynamics within the mold cavity that determine the molded part quality. As previously mentioned, these transfer functions are determined by the laws of physics. For this reason, the quality dynamics have been explicitly separated from the molding machine. Wang (1989), Marchal (1992), and Turng (1995) have developed sophisticated numerical simulations that accurately describe many of the dynamics of polymer processing. However, these simulations are generally not of the form shown in Figure 1 and are inconvenient for exploring the design requirements for intelligent control for processing of polymeric materials. For this reason, a simple model has been developed to provide an explicit investigation of a few aspects of the injection molding process. The block diagram shown in Figure 2 represents a subset of the plant from Figure 1. In the diagram, there are two inputs, u and C, which are converted through the machine and quality dynamics to three outputs, Q_dimension, Q_flash, and Q_short.