Method of Tuning and Automatic Regulation for Injection Molding

Transcription

1 Method of Tuning and Automatic Regulation for Injection Molding Kourosh Danai David Kazmer University of Massachusetts Amherst Abstract Efficient tuning of injection molding has been an enduring problem in industry due to the lack of accurate relationships between the machine inputs and molded part quality. These relationships have not been developed, and may never be developed, due to the complex properties of thermoplastic materials coupled with the complex pressure and temperature dynamics of the injection molding process. The objective of this research is to develop an efficient and robust method of tuning and regulation for the injection molding process so that the desired quality of injection molded parts can be readily achieved and consistently ensured for commercial production. This paper reports the progress of this research. 1 Introduction The demand for production of complex parts, required for improved functionality and ease of assembly, has necessitated injection molded parts with tighter tolerances and superior finish. This, in turn, has increased demands for more accurate control of the process [1]. Towards this objective, significant progress has been made in improving the stability of the process on-line [2-5]. However, relatively little attention has been paid to efficient specification of setpoints for various machine inputs such as melt and mold temperatures, injection pressure, and injection speed. Ideally, these setpoints should be specified so as to produce parts with acceptable part quality attributes, which for an injection molded part would typically be size, surface topography, and/or mechanical properties (e.g., tensile strength, flexural strength). The primary difficulty in selecting appropriate setpoints in injection molding is the complexity of the process and its variability. Polymers have extremely complex material properties: non-newtonian, non-isothermal rheology together with highly temperature and pressure dependent thermal properties. During processing, the material undergoes temperature and pressure increases and significant shear deformation, followed by rapid decay of temperature and pressure in the cavity, leading to solidification and locking of residual stress, orientation, and other part properties that determine the molded part quality. In addition to process complexity, process variability poses an impediment to maintaining the desired part quality during production. A major source of product inconsistency is variation in material properties. For instance, small changes in viscosity, density, or composition may occur when one material is substituted by another having similar flow properties, regrind is mixed with virgin material, a material is used after it has been stored over an extended period of time, or a switch is made between different batches of the same material grade. A second source of variability involves the process machinery. For instance, molding machines of different injection cylinder and clamp design will have very different machine dynamics, providing different levels of molded part quality for the same process set-points -- even `identical' molding machines from the same manufacturer can induce significant quality variation due to differences in machine controllers as well as varying amounts of wear in the melt and hydraulic delivery systems. Other sources of variability stem from human interaction and the physical environment in which the molding machine is operating (e.g., outdoor temperature may limit the

2 effectiveness of evaporative coolers which may change the temperature of the plant water; humidity can affect the dryness of the materials entering the barrel which may introduce severe quality inconsistencies). The traditional approach to machine input selection (tuning) in the plastics industry has been `trial and error'. For this, shots are taken during start-up and part quality attributes are measured after each shot to evaluate the quality of produced parts. A human expert then uses his/her knowledge of the process to select the machine inputs in such a way as to improve the quality of the part from shot to shot. This tuning exercise is repeated until the specifications for part quality are satisfied. The main drawback of the traditional tuning approach is its inefficiency due to its `ad hoc' nature. Humans usually use linear relationships to relate machine inputs to quality attributes, so they often have difficulty adjusting the inputs over large ranges [6]. They also tend to treat the various attributes as independent, thus, ignore the couplings among the attributes. This and the added difficulty of coping with noise often lead to time-consuming tuning sessions and considerable waste. An alternative to the traditional trial and error approach has been the use of `expert systems'. Troubleshooting expert systems, which have attracted considerable attention in recent years, represent corrective guidelines in the form of `if-then' rules [7-9], so they have the appeal of replacing the human expert in providing trouble-shooting knowledge. Expert systems, however, are limited by their non-quantitative nature and their inability to cope with quality issues not addressed by the rules. A more methodical approach to tuning injection molding processes is Design of Experiments (DOE), where an empirical model is formed based on data obtained from a set of designed experiments [1]. Based on this model, the objective function of an unconstrained optimization problem is defined as a function of the part quality attributes, and the set of inputs that produce the best quality attributes are obtained as the `optimal' point of this optimization problem. While DOE-based methods offer a systematic approach to tuning that can also be used for mold qualification [11], they are only practical for large scale injection molding applications where the high cost associated with constructing a comprehensive empirical model can be justified. In this research a method of tuning is considered for injection molding that incorporates aspects of both the trial and error approach and DOE. This method, analogous to the trial and error approach, uses measurements of part quality attributes to evaluate the inputs. It differs from this approach, however, in that like DOE methods it uses an input-output (I-O) model to select the inputs. In this method, the inputs are applied to the process only when the search for them has been exhausted based on the current I-O model. As such, this method uses the I-O model as a `virtual' process and does not require feedback from the process for each iteration of the search. The Virtual Search Method (VSM) [12] refers to the process to (1) test the feasibility of the inputs obtained from the current I-O model and (2) to update the I-O model using the measurements of part quality attributes obtained from the process. According to this scheme, the I-O model is updated only when it no longer provides guidance towards the feasible region. 2 Virtual Search Method The block diagram of VSM is shown in Fig. 1. It consists of a `search algorithm' that determines prospective changes to the machine inputs for the next part, an `input-output (I-O) model' which estimates the corresponding changes to the part attributes, and a `learning algorithm' to update the I-O model after each cycle based on part quality measurements [12].

3 Molding Process Molded Attributes Learning Algorithm Input- Output Model Speculated Attributes Search Algorithm Figure 1: Schematic of the Virtual Search Method 3 Status of the Project Research during the first year of this project was focused on theoretical development of VSM and its experimental validation. For its theoretical development, test beds were devised to simulate various aspects of injection molding with different degrees of process nonlinearity. 3.1 Theoretical Development Using simulation, several facets of VSM s operation were studied. The following are the conclusions reached from these studies: Convergence needs to be defined more formally for VSM, if it is to be used for process optimization as well. Presently, VSM is declared to have converged as soon as one good part is produced (the process output is within its feasible region). However, as training continues in the hope of improving the quality of this output (towards an optimal point), bad parts may be produced because of the mismatch between the feasible regions of the I-O model and the process. This points to the inadequacy of the current criterion used for convergence. The criterion for learning needs to be redefined so as to improve the overlap between the feasible regions of the I-O model and the process. Currently, the I-O model is being trained to replicate the outputs of the process. The basic assumption is that this strategy will lead to the convergence of its feasible region as well. The results obtained from simulation indicate that while the present training strategy increases the likelihood of producing good parts (producing outputs within the process feasible region), it does not necessarily continually improve the overlap of the I-O model s feasible region with that of the process.

4 A Criterion needs to be established to assess the suitability of the I-O model during the first stages of VSM operation. The suitability of the I-O model s ability to ultimately represent the process is always a concern during the application of VSM. Availability of a criterion to assess the suitability of the I-O model will be beneficial during VSM s application. Sometimes a hybrid VSM, consisting of both linear and nonlinear I-O models is more effective. For these cases, a criterion needs to be established to evaluate the suitability of the recommendations from each model. Given that sometimes a linear model is incapable of representing the process, it is advantageous to train simultaneously models with varying degrees of complexity from the start of the tuning operation. This then required evaluating the recommended modifications to the process from each model with the objective of using one of them or their combination. Theoretical development of VSM will be continued by addressing the above issues. 3.2 Experimental Development In order to ascertain its practical utility in stringent production settings and to determine the developmental issues that need to be adderssed, VSM was tested at GE Plastics in tuning for production of digital video disks (DVD) molded of a commercial, optical grade polycarbonate. The DVD geometry consists of a center gated disk of radius 6 mm and thickness.6 mm. Because it is used in a high density optical storage application, the production requirements include minimization of residual stress, birefringence, flatness, and repeatable track-track dimensional consistency (groove replication). The VSM was utilized to tune six processing parameters on a Sumitomo SD3 molding machine. The input parameters included melt barrel temperature, mold coolant temperature, mold coolant differential between the moving and stationary halves, polymer injection velocity, first stage clamp tonnage, and first stage clamp tonnage delay. The range for each parameter was set to the global processing limits conventionally used for optical molding. The production of optical media is not always possible at all input levels. The four output parameters consisted of minimum and maximum birefringence levels across the molded DVDs, maximum tangential deviation, and maximum radial deviation from a flat surface. The experimental methodology consisted of allowing the melt and mold temperatures to reach steady state, after which twenty discs were molded and discarded. The next five samples were collected, allowed to cool, and then analyzed with a specialized optical media measurement instrument (Dr. Sehnck Prometheus, Germany). Altogether, three days of testing were conducted during which approximately eight thousand discs were produced. The changes made to the input and the variation of the outputs during the application of VSM are shown in Figures 2 and 3, respectively. The results in figure 3 indicate that a satisfactory set of inputs was achieved at the sixth iteration. However, tuning was continued (including training the I-O model) so as to test the robustness of VSM. As indicated by the outputs in Figure 3, continued training of the I-O model increases the likelihood of finding a satisfactory set of inputs, however, it does not guarantee convergence at each iteration. Some of the theoretical issues listed above aim at improving the robustness of VSM.

5 Output 3 Input Input 1 35 Input Input Input Input Figure2. Changes made to the inputs by VSM during tuning. Input 1: melt barrel temperature, Input 2: mold coolant temperature, Input 3: mold coolant differential between the moving and stationary halves, Input 4: polymer injection velocity, Input 5: first stage clamp tonnage, and Input 6: first stage clamp tonnage delay Output Output Output Figure 3. Variation of outputs during tuning by VSM. Output 1: minimum birefringence levels across the molded DVDs, Output 2: maximum birefringence levels across the molded DVDs, Input 3: maximum tangential deviation, and Output 4: maximum radial deviation from a flat surface.

6 Acknowledgments This research is supported by NSF Grant No. DMI The collaboration of GE Plastics in the experimental phase of this research is acknowledged. References [1] Agrawal, A. R., Pandelidis, I. O., and Pecht, M., 1987, Injection-Molding Process Control: A Review, Polymer Engineering and Science, Vol. 27, pp [2] Chiu, C. P., Wei, J. H., and Shih, M. C., 1991, Adaptive Model Following Control of the Mold Filling Process in an Injection Molding Machine, Polymer Engineering and Science, Vol. 31, p. 15. [3] Gao, F., Patterson, W. I., and Kamal, M. R., 1994, Self-tuning Cavity Pressure Control of Injection Mold Filling, Advances in Polymer Technology, Vol. 13, p [4] Grolman C. P., and Nunn, R. E., 1991, Adaptive Process Control for Injection Molding, J. of Reinforced Plastics and Composites, Vol. 9, p [5] Kazmer, D. O., and Barkan, P., 1996, Multi-cavity Pressure Control During the Filling and Packing Stages of Injection Molding, Polymer Engineering and Science, [6] Moray, N., Lootsteen, P., Pajak, J., 1986, Acquisition of Process Control Skills, IEEE Trans. Syst., Man, and Cybernetics, Vol. SMC-16, p [7] Rogers, J. K., 1991, Intelligent Molding: Expert Systems are Coming on Line Now, Modern Plastics, Vol. 68, pp. 56. [8] Kameoka, S., Haramato, N., and Sakai, T., 1993, Development of an Expert System for Injection Molding Operations, Advances in Polymer Technology, Vol. 12, No. 4, pp. 43. [9] Farrell, R. E., and Dzeskiewicz, L., 1994, Expert System for Injection Molding, SPE ANTEC Conference Proceedings, San Francisco, CA, pp [1] Schmidt, S., and Launsby, R., 1988, Understanding Industrial Designed Experiments, Air Academy Press, Colorado Springs, Colorado. [11] Martin, M. F., Bontumasi, F., and Young, G., 1995, Practical Application of DOE in the Total Quality Injection Molding Process, SPE ANTEC Conference Proceedings, Boston, MA, pp [12] Ivester, R., Danai, K., Automatic Tuning and Regulation of Injection Molding by the Virtual Search Method, Journal of Manufacturing Science and Engineering, 12 (5), p. 323 (1998).