Int J Adv Manuf Technol (2014) 72:765–777 DOI 10.1007/s00170-014-5711-0
ORIGINAL ARTICLE
A nondestructive online method for monitoring the injection molding process by collecting and analyzing machine running data Peng Zhao & Huamin Zhou & Yong He & Kan Cai & Jianzhong Fu
Received: 19 June 2013 / Accepted: 10 February 2014 / Published online: 25 February 2014 # Springer-Verlag London 2014
Abstract Nondestructive online monitoring of injection molding processes is of great importance. However, almost all prior research has focused on monitoring polymers in molds and damaging the molds. Injection molding machines are the most important type of equipment for producing polymeric products, and abundant information about actual polymer processing conditions can be obtained from data collected from operating machines. In this paper, we propose a nondestructive online method for monitoring injection molding processes by collecting and analyzing signals from injection molding machines. Electrical sensors installed in the injection molding machine, not in the mold, are used to collect physical signals. A multimedia timer technique and a multithread method are adopted for real-time large-capacity data collection. An algorithm automatically identifies the different stages of the molding process for signal analysis. Moreover, ultrasonic monitoring technology is integrated to measure the cavity pressures. Experimental results show that our nondestructive method can continuously monitor the injection molding process in real time and automatically identify the different stages of the molding process. The packing parameters, including the filling-to-packing switchover point and the packing time, can be optimized based on these data. Furthermore, the ultrasonic reflection coefficient and the actual cavity pressure have similar trends, and our technique for measuring the cavity pressure is accurate and effective. P. Zhao : Y. He : K. Cai : J. Fu (*) The State Key Lab of Fluid Power Transmission and Control, Zhejiang University, Hangzhou 310027, Zhejiang, People’s Republic of China e-mail:
[email protected] H. Zhou The State Key Lab of Material Processing and Die & Mould Technology, Huazhong University of Science and Technology, Wuhan 430074, Hubei, People’s Republic of China
Keywords Injection molding . Process monitoring . Nondestructive . Stage identification . Cavity pressure measurement . Packing parameters
1 Introduction Injection molding is by far the most important process for producing polymeric products. Today, more than one third of all polymeric products are produced by the injection molding process [1]. However, injection molding is a highly complicated batch process. Polymeric granules are first melted and then forced into a closed mold where the molten polymers solidify to form the final product, and the polymers in the cavity undergo a complex thermomechanical process [2]. The qualities of the final product are affected by the molding process to a great extent. Improper settings of process variables will produce various defects in the final product [3–5]. Monitoring the injection molding process online is of great importance because it can help engineers understand the entire molding process and optimize the process to consistently produce high-quality products. In the last few decades, researchers have employed various destructive methods for online monitoring of the injection molding process including temperature and pressure sensors [6–9], visible mold detectors [10–13], capacitive transducers [14–16], fluorescent sensing [17], and near infrared spectroscopy [18, 19]. During the injection molding process, molten polymers are formed and solidified in closed molds surrounded by mold steels. Almost all of the prior research focused on the polymers in the mold, and sensors were installed in the molds to contact with the polymers. The molds were damaged by those methods. Temperature and pressure sensors are widely used to monitor polymers, and holes are drilled into the molds to accommodate them. For a visible
766
detector, a glass window is inserted in the mold to observe the injection molding process. For capacitive transducers, two electrodes are placed in contact with the polymers to collect cavity information online. Similarly, fluorescence and near infrared spectroscopy both require embedded optical fiber sensors to obtain images inside the cavity for further analysis. It is obvious that these destructive methods destroy the molds, harm the appearance of the final product, and reduce the mold’s working strength, especially for high-processing pressures. In addition, these destructive methods are inconvenient and expensive. Hence, these methods are inappropriate for industrial applications. Injection molding machines are the most important type of equipment for producing polymeric products. Data collected from operating machines contain abundant information about actual processing conditions during polymer molding [20]. Polymeric granules are heated and melted in the barrel of the injection unit. The temperature of the polymers in the cavity is influenced by the barrel temperature, the back pressure, etc. Dontula et al. proposed a theoretical model for the effect of the machine variables on the melt temperature [21]. Molten polymers are forced into a cavity with a screw by an injection hydro-cylinder. The pressure and fill velocity of the polymers in the cavity are affected by the oil pressure and the screw position in the injection unit, respectively. Wang et al. established an integral mathematical model for the relationship between the packing pressure and the oil pressure in the injection hydro-cylinder in a servo motor-driven injection molding machine [22]. Dubay et al. described the relationship between the fill velocity and the screw velocity and presented two predictive controllers for the screw velocity [23]. Consequently, the running data of injection molding machines can provide valuable information about the injection molding process, and it is a very valuable ideal for nondestructive online monitoring of the injection molding process by collecting and analyzing signals from the injection molding machines. In this paper, a nondestructive method is proposed for online monitoring the injection molding process. Traditional pressure, temperature, and displacement sensors are installed in the injection molding machine, not in the mold, for collecting data while the machine runs. A multimedia timer method and a multithread technology are adopted for real-time large-capacity data collection. An algorithm automatically identifies the different stages of the injection molding process based on the stroke curves. This proposed nondestructive method is an open platform. An ultrasonic monitoring technique is integrated for measuring the cavity pressures by fusing the machine’s running data and the ultrasonic signals. In general, this proposed monitoring method offers several advantages: (1) nondestructive, (2) real time, (3) large capacity, (4) wide open, (5) low cost, and (6) health and safety.
Int J Adv Manuf Technol (2014) 72:765–777
2 Experimental 2.1 Machine signal collection technique The hydraulic reciprocating screw injection molding machine is the most widely used type of equipment in the polymeric industry due to its reliability and overall performance [24]. Hence, the hydraulic reciprocating screw injection molding machine is the subject of this paper. A simplified general layout for a method to monitor an injection molding machine is illustrated in Fig. 1. Traditional electrical sensors such as thermocouples and displacement and pressure probes are installed in the injection molding machine, not in the mold, for collecting the pressures, positions, temperatures, and time during molding. The pressure signals are measured with pressure probes. The position signals are monitored with resistive electronic sensors, and the temperature signals are detected with K-type thermocouples. Time signals are recorded according to the clock system of a supervising computer. These signals can be divided into three categories: analog, discrete, and temperature signals. Three data collection cards, the USB4716, the USB-4750, and the USB-4718 (all made by Advantech Co., Ltd.), are used to pretreat and transform the analog signals, the discrete signals, and the temperature signals to digital signals, respectively. Then, these digital signals are read and saved by a supervising computer that also serves as a data processor. A multimedia timer method and a multithread technique are employed for real-time largecapacity data collection, and they are explained in the following sections. 2.1.1 Real-time data collection The injection velocity of an injection molding machine is often as high as 0.1 mm/ms. To ensure accuracy, the sampling period of the monitoring method should be 1 ms, i.e., 1,000 data per second. In the Windows platform, there are two main real-time data acquisition methods, the interrupt technique and the software technique. As shown in Fig. 1, there are three data collection cards in this monitoring method. Obviously, the interrupt technique cannot synchronize the data from these three data collection cards. The signals acquired by the interrupt technique from different cards cannot be placed in the same time sequence. Conversely, the software technique can be employed to acquire signals by traversing the ports of these four cards to synchronize the data collection. The WM_TIMER timer and the multimedia timer are two types of timers used with the Windows operating system. The Windows operating system is a message-driven, nonpreemptive priority multitasking operating system. The maximum accuracy of the WM_TIMER timer is 55 ms. In contrast, the multimedia timer is activated by a hardware interrupt
Int J Adv Manuf Technol (2014) 72:765–777
767
Fig. 1 A simplified general layout of a method for monitoring an injection molding machine
with the highest priority. Moreover, the multimedia timer does not send any messages, and there is an independent thread for executing the callback functions. The accuracy of the multimedia timer is 1 ms. Hence, the multimedia timer is employed as a data collection timer in this paper. 2.1.2 Large-capacity data collection Injection molding machines often operate all day and all night, so the monitoring method should record signals continuously and remain stable for a long time. Moreover, the recorded data should be saved for playback and off-line analysis. A singlethreaded program can hardly achieve these purposes. Hence, a multithreaded program is designed, and the three threads run in parallel. These three threads are the main thread, the data acquisition thread, and the data storage thread. (a) The main thread. The main thread processes documents and views, interactions with users, and starts or stops the other threads. (b) The data acquisition thread. When users click the command to start
data acquisition, this thread begins. According to the preset parameters, this thread acquires signals from the ports of the data collection cards in real time, and then saves these signals and the time as data in the computer memory. The structure of the collected data is a list, an ordered collection of values that can be implemented easily. The main thread displays the data from the list. When users click the command to stop data acquisition, this thread ends immediately. (c) The data storage thread. This thread begins after the data acquisition thread is started. The data storage thread transfers the collected data from the memory to a file on the computer hard disk. At the same time, these data are deleted from the list after they have been displayed by the main thread. When the user stops the data acquisition, this thread ends after the list of collected data is empty. In multithreaded programming, the most important problem is the communication and the synchronization of the main thread, the data acquisition thread, and the data storage thread. The flowchart of these three threads for acquiring signals is shown in Fig. 2. The main thread takes charge of starting and
768
Int J Adv Manuf Technol (2014) 72:765–777
Fig. 2 The flowchart of the three threads for collecting signals
stopping the other two threads. The communication and the synchronization of the different threads are based on shared memory. The data acquisition thread places signals in the shared memory in a list format. The main thread displays the signals from the shared memory. The data storage thread saves the signals from the shared memory in the computers’ hard disk and deletes them from the shared memory. In this manner, the acquired signals are saved in the hard disk and not in the memory. In theory, if the computer hard disk is large enough, then the monitoring process can continue indefinitely.
by the resistive electronic sensors. An example of the original data collected for an injection stroke curve is displayed in Fig. 3. Nearly all of the filtering algorithms, such as median filtering, low-pass filtering, and moving average filtering, cannot remove the noise from these stroke curves and identify the key points. In fact, every stroke curve acquired by a resistive electronic sensor is a set of points, which are measured at 1,000 data per second in this work. Thus, the process of searching for the key points can be divided into the following two steps.
2.2 Machine signal analysis technique
Step 1 Search for potential key points using an area criterion. A stroke curve is composed of a set of points pi (i=1, 2,…, n). As shown in Fig. 4, the area of A(p1, pm) bounded by the connection p1pm and the curve p1, p2,…,…, pm (m
As mentioned above, injection molding machines often work all day and all night, and huge amounts of data can be collected and analyzed. Injection molding is a typical batch process. Identifying the stages online is crucial for analyzing the injection molding process. In this section, we propose a stage identification algorithm based on the stroke curves collected by the resistive electronic sensors. The key problem for the stage identification algorithm is to automatically find key points (intersections of different stages) in the stroke curves. In theory, if no noise existed, then these stroke curves would be straight line segments. However, there is a great deal of irregular noise in the stroke curves acquired
Int J Adv Manuf Technol (2014) 72:765–777
769
Fig. 3 An original set of data collected for an injection stroke curve
for handling points distributed widely in a narrow zone. Second, the total calculation is small. Each point is calculated only once, including two multiplications and two additions and subtractions. Aðp1 ; pm Þ ¼
1 ! ! ! ! ð p p p1 p3 þ p1 p3 p1 p4 2 1 2 ! ! þ … þ p p p p Þ ð1Þ 1 m−1
2.3 An extension: integrating ultrasonic monitoring technology This proposed nondestructive monitoring method is an open platform, and other nondestructive technologies can be integrated to obtain more information about the injection molding process. In this section, an ultrasonic monitoring technology is described as an example.
1 m
2.3.1 Brief introduction to ultrasonic monitoring technology Step 2 Identify the key points from the set of potential key points by an angle criterion. If all of the potential key points are connected by lines, then there is a series of line segments. As shown in Fig. 5, the points Qi−1, Qi, Qi+1, and Qi+2 are potential key points, and the angles αi−1 and αi are the azimuthal angles for potential key points Qi−1 and Qi, respectively. Δαi is the absolute value of the difference between αi−1 and αi. The calculation formula for αi and Δαi is shown as Eq. (2), where the subscripts x and y denote the values of the x- and the y-coordinates of the potential key point, respectively. The value of Δαi can be used to identify the key points. If Δαi is bigger than a predefined threshold angle Tangle, indicating a sharp turn, then the potential key point Qi is a key point. Otherwise, Qi is not a key point and i=i+1. h . i αi ¼ arctg Qðiþ1Þy −Qiy Qðiþ1Þx −Qix Δαi ¼ jαi −αi−1 j
Fig. 4 The schematic diagram of the area A(p1, pm)
Ultrasounds are mechanical waves, which can easily transmit into mold steels and return rich information about polymers. More recently, ultrasonic technology has shown great potential as a nondestructive test method for characterizing the polymer processing process [25–32]. Michaeli and Starke found that ultrasonic signals can supply information for much longer periods of time than, for example, pressure signals [26]. He et al. employed ultrasonic technology to monitor the injection molding process in an attempt to establish a fundamental understanding of the processing/morphology/ultrasonic signal relationships [27]. Ono et al. successfully monitored information about the filling, the solidification, the shrinkage, and the detachment of polymer material inside the mold cavity using ultrasonic sensors [29, 30]. However, the above studies involved qualitative research, and they only suggested the possible potential of ultrasonic technology for monitoring molding processes. Michaeli and
ð2Þ
Fig. 5 The schematic diagram of the angle Δαi
770
Starke noted that the most important limitation is the lack of knowledge about ultrasonic signals, which depend strongly on the polymer processing conditions [26]. Although ultrasonic signals contain rich information, it is impossible to quantitatively measure an injection molding process without collecting information about the polymer processing conditions. If some of the destructive methods discussed in Sect. 1 are used to acquire the processing conditions, then the most important advantage of ultrasonic technology, i.e., its nondestructive nature, is lost. 2.3.2 Cavity pressure measurement This proposed nondestructive monitoring method can be adopted to acquire processing conditions. Hence, the integration of this proposed method and the ultrasonic monitoring technique is very helpful for quantitatively characterizing the key processing parameters such as the cavity pressure. The cavity pressure is very critical for achieving high-quality products. The cavity pressure determines the evolution of the conditions of the polymer inside the mold cavity, and it has very strong influence on the quality of the final products, particularly their dimensions, dimensional stability, mechanical behavior, and surface quality [33]. Nondestructive online measurements of cavity pressures are significant for understanding and optimizing the molding process. However, the nondestructive measuring of the cavity pressure in real time remains a challenge, and very few methods are available. It is well known that the pressure-volume-temperature (P-V-T) property is a very important physical parameter for polymers, which describes the relationships between the pressure, the density (the reciprocal of the specific volume), and the temperature [8]. The P-V-T property of a given polymer is constant. Hence, the cavity pressure is a function of the polymer density and temperature. Based on this interior P-V-T property, we present a nondestructive approach to measure the cavity pressure by the fusion of data from a running machine and ultrasonic signals. The density of a polymer material is characterized by ultrasonic signals. An ultrasonic transducer is placed in contact with the outside surface of the mold. A schematic diagram of ultrasonic technology, which is based on pulsed ultrasound
Fig. 6 The schematic diagram of the ultrasonic method for measuring the cavity pressure
Int J Adv Manuf Technol (2014) 72:765–777
Fig. 7 The structure of the NN model for measuring the cavity pressure
and performed in reflection mode, is described as Fig. 6. The ultrasonic transducer generates pulsed ultrasounds and receives the ultrasounds reflected by the mold/cavity interface. The mold is made of NAK80 steel, while the cavity is filled with polymer. According to the ultrasonic theory, if there is ultrasound transfer at the interface between two different media, some of the ultrasound is reflected back from the interface and the rest is transmitted. The reflection coefficient R is defined as Eq. (3) R¼
zmold −zpolymer zmold þ zpolymer
ð3Þ
where Zmold and Zpolymer are the acoustic impedances of NAK80 steel and the polymer, respectively. These impedances can be calculated by Eq. (4) zmold ¼ ρmoldcmold zpolymer ¼ ρpolymer cpolymer
ð4Þ
where ρmold and ρpolymer are the densities of NAK80 steel and the polymer material, respectively, while cmold and cpolymer are the ultrasound velocities for NAK80 steel and the polymer, respectively. Substituting Eq. (4) into Eq. (3), the ρpolymer can be given by Eq. (5). As shown as Eq. (5), the polymer density is highly influenced by the reflection coefficient R of the ultrasonic signals. Thus, the reflection coefficient can be used to characterize the polymer density. ρmold cmold 1−R ρpolymer ¼ ð5Þ 1þR cpolymer
Fig. 8 The apparatus for the data collection system
Int J Adv Manuf Technol (2014) 72:765–777
771
Fig. 9 A set of data collected by this monitoring system
The mold temperature and the injection hydro-cylinder pressure are characterized by the data collected from a running machine through our nondestructive monitoring method. We assume that the polymer temperature is the same as the mold temperature at the mold/cavity interface because the polymer and the mold are in close contact during processing. Therefore, the temperature of the polymer in the cavity is monitored by a K-type thermocouple. This thermocouple is not embedded in the mold, but it is only used to measure the temperature of the mold at the ultrasonic measurement point. In addition, the injection hydro-cylinder drives the screw and the molten polymers into the cavity during processing. The cavity pressure is affected significantly by the pressure in the hydro-cylinder. Hence, the pressure in the hydro-cylinder is also recorded by a pressure probe, which is installed in the cylinder. A soft computing method based on a feed-forward neural network (NN) is used to analyze the polymer density, the polymer temperature, and the injection hydro-cylinder pressure. The P-V-T relationship contains dozens of uncertain variables for modeling nonlinear relationships among pressure, density, and temperature, and it is difficult to obtain the P-V-T relationship under normal processing conditions [8]. Moreover, the P-V-T relationship varies greatly among different polymers. Hence, it is difficult to directly calculate the cavity pressure from the P-V-T relationship. The NN is one of the most widely used models for modeling nonlinear functional relationships. It has been proved that a NN can simulate any exact continuous function [34, 35]. In this paper, a feed-forward NN is constructed to imitate the complicate relationships among pressure, density, and temperature for measuring cavity pressure, as shown in Fig. 7. The inputs
of the NN are the ultrasonic reflection coefficient, the mold temperature, and the injection hydro-cylinder pressure. The reflection coefficient is calculated by the amplitudes of the incident ultrasonic signals and those of the reflected ultrasonic signals. These ultrasonic signals are first denoised by a low-pass filter, and then, their amplitudes are obtained by a fast Fourier transform (FFT) at the frequency of the ultrasonic transducer. As discussed previously, the mold temperature equals the polymer temperature at the mold/cavity interface, and the injection hydrocylinder pressure significantly affects the cavity pressure. Hence, these parameters are used as inputs to improve the performance of the NN. The number of nodes in the hidden layer is 30. The output of the NN is the cavity pressure.
3 Results and discussion The apparatus for the monitoring system is shown in Fig. 8. Three experiments are discussed in this section. Table 1 The stages that correspond to the key points Key points
Represented stages
Key points
Represented stages
A B C D E
Mold closing start point Mold closing end point Injection start point Injection end point Plasticization start point
F G H I J
Plasticization end point Mold opening start point Mold opening end point Ejection start point Ejection end point
772
Int J Adv Manuf Technol (2014) 72:765–777
Fig. 10 The photograph of experimental mold and its molded parts
3.1 Characterization of injection molding process In this experiment, our process monitoring system is applied to collect and analyze the information collected while an injection molding machine runs. The injection molding machine is an HTL160 machine (made by Ningbo Haitai Plastic Machinery Co., Ltd), which is fully hydraulic and operates automatically. K-type thermocouples (WRNT07/1.0 M made by Xinxing Co., Ltd.), pressure probes (HM181X210VSV00 made by Bosch Rexroth China), and resistive electronic sensors (LWH-300 made by Novotechnik Germany, Inc.) are installed in the HTL160 machine. The monitoring system collects the signals from the HTL160 without interruption for 5 days. The number of collected signals is nine. The size of the file saved in the hard disk is 3,856.2012 MB. A computer hard disk with a capacity of 250 GB can, in theory, record nine signals without interruption for more than 330 days. Hence, this monitoring system can collect signals for a long time. Moreover, the time interval between two consecutive data is 1 ms, and this monitoring system can collect signals in real time. The injection molding process is characterized by the signals collected from the injection molding machine. A set of collected data is shown in Fig. 9. The hydro-cylinder pressures (the carriage and the clamping hydro-cylinder pressures) are acquired by pressure probes, while the stroke curves (the ejection, the clamping, and the injection strokes) are obtained with resistive electronic sensors. As shown in Fig. 9, the stage
identification algorithm identifies several key points (marked by "+") on the injection, the clamping, and the ejection stroke curves. These key points, especially the points identified with letters (A, B, C, etc.), represent different stages of the injection molding process. The stages that correspond to different key points are listed in Table 1. Region (A→B) provides information about the mold closing stage. At point A, higher pressure is needed in the hydro-cylinder to overcome the static friction in the clamping unit. At point B, higher hydraulic pressure is also required to tightly clamp the mold. The two key points between points A and B represent the three stages of mold closing. Region (B→C) stands for the advancing stage of the injection unit. Region (C→D) provides information about the injection stage. In this stage, molten polymers are injected through the nozzle into the mold cavity, and a reactive force is imparted to the injection unit as the nozzle leaves the mold. Hence, high pressure is required in the carriage hydro-cylinder to overcome the reactive force and make the nozzle close the mold’s gate. Regions (D→E) and (E→F) indicate the cooling and the plasticization stages, respectively. Because of leakage in the one-way valve, there is a small pressure loss in the carriage hydro-cylinder between points C and F. Region (F→G) denotes the return stage of the injection unit. Region (G→H) represents the mold opening stage. In this stage, the “shock of mold open” phenomenon occurs. The pressure in the clamping unit increases sharply because the oil cannot flow out of the clamping hydrocylinder unit quickly enough during the mold opening stage. Region (I→J) is the ejection stage, and the molded products are ejected. Thus, a complete cycle of the injection molding process is finished, and the next cycle is ready to start. 3.2 Optimization of the injection molding parameters This experiment includes practical applications for optimizing packing parameters such as the filling-to-packing switchover point and the packing time. The injection molding machine is an HTL160 made by Ningbo Haitai Plastic Machinery Co., Ltd. The experimental mold and its molded parts are shown in
Fig. 11 The experimental data at different screw position switchover points. a 32 mm. b 38 mm
(a) Screw position at 32 mm
(b) Screw position at 38 mm
Int J Adv Manuf Technol (2014) 72:765–777
773
Fig. 12 The comparison of the injection stroke positions and the product weight for different packing times
Fig. 10, and the polymer is poly styrene (PG-33 from Zhenjiang Chimei Petroleum Chemical Co., China). The filling-to-packing switchover point refers to the position at which the injection molding machine switches from the filling stage to the packing stage, and it plays a crucial role in ensuring the quality of the molded products [36–38]. Early switchover induces incomplete filling, i.e., short shot, while late switchover results in excessive filling, i.e., flash [14]. In practical production situations, the screw position switchover is the most widely used method. The filling stage would ideally finish when the molten polymer reaches the end of the mold cavity. However, unknown cavity volumes, leaking nozzles and the compressibility of molten polymers can easily trigger a machine to switch over, thereby causing incomplete or excessive filling [38]. Our monitoring method can be applied to optimize the filling-to-packing switchover point.
Fig. 13 The cross-sectional view of the experimental mold
The injection stroke signals and the injection hydro-cylinder pressure signals are collected simultaneously, and the fillingto-packing switchover point is identified easily. Two experimental data at different screw position switchover points are shown in Fig. 11. It is important to note that the original point for the screw is the nozzle, i.e., the screw position continuously decreases during the filling stage. In this experiment, a skilled machine operator set the switchover point at 32 mm according to his experience, and the experimental result is displayed in Fig. 11a, which shows excessive filling characterized by a pressure peak. This pressure peak continues to force an excessive amount of molten polymer into the cavity until the switchover, and it adds weight and stress to the molded products. As shown in Fig. 11a, the hydraulic pressure increases sharply at the screw position of 38 mm, which indicates that the cavity is filled completely. Therefore, the optimum filling-to-packing switchover point is at 38 mm, and this experimental result is shown in Fig. 11b. The filling stage smoothly switches to the specified packing pressure, and there is no pressure peak, which demonstrates that 38 mm is the optimum switchover point. The packing time is a parameter that represents the duration of the packing stage, and it is very important in the injection molding process because it not only influences the final product quality but also determines the total processing time [39]. However, in production, setting the packing time is considered to be a “black art,” which relies heavily on the experience and the knowledge of the machine operator and requires a trialand-error process. This trial-and-error method is timeconsuming, and there is no assurance that the optimum packing time will be determined. In general, the packing stage is effective until the gate solidifies and the molten polymers cannot be forced into the cavity to compensate. Hence, the optimum packing time is the gate freeze-off time. In this
774
experiment, the packing time (i.e., gate freeze-off time) is optimized by monitoring the variations of the injection stroke
Int J Adv Manuf Technol (2014) 72:765–777
curve, and the optimum packing time is 2.5 s. To verify this optimization result, the injection stroke positions and product
Fig. 14 A set of experimental data for a the mold temperature, b the injection hydro-cylinder pressure, c the ultrasonic reflection coefficient, and d the actual cavity pressure
(a) Mold temperature
(b) Injection hydro-cylinder pressure
(c) Ultrasonic reflection coefficient
(d) Actual cavity pressure
Int J Adv Manuf Technol (2014) 72:765–777
weights at different packing times are compared in Fig. 12. From this figure, it can be observed intuitively that the injection stroke changes little after 2.5 s, and similarly, the product weight does not change at the same time, which indicates that the gate is solidified and no more polymer is injected into the cavity. Hence, with the help of the injection stroke curve collected by our monitoring method, the packing time can be optimized quickly to produce high-quality products at a minimal cost. 3.3 Verification of the cavity pressure measurement The structure of the experimental mold is shown in Fig. 13. The experimental cavity is a flat sheet, 80×140 mm and 4-mm thick. A 6190A cavity pressure/temperature probe (made by Kistler Instruments Ltd.) and a 5-MHz longitudinal wave pulsing/receiving ultrasonic transducer (made by Shantou Institute of Ultrasonic Instruments Co., Ltd.) are used to collect the cavity pressure and the ultrasonic signals, respectively, at the same place. Their locations are shown in Fig. 13. The experimental polymer is polyvinyl chloride (PVC-SG5 from Xinjiang Tianye (Group) Co., Ltd.). A set of experimental data for the mold temperature, the injection hydro-cylinder pressure, the ultrasonic reflection coefficient, and the actual cavity pressure are shown in Fig. 14. A low-pass filter is applied to process the ultrasonic reflection signal. From this figure, it can be seen intuitively that the hydro-cylinder pressure is quite different from the actual cavity pressure, and it cannot be used to characterize the cavity pressure. In contrast, the ultrasonic reflection coefficient and the actual cavity pressure share similar trends. As discussed in Sect. 2.3.2, a feed-forward NN is developed for online measurements of the cavity pressure. The inputs of the NN are the mold temperature, the injection hydro-cylinder pressure, and the ultrasonic reflection coefficient, while the output of the NN is the cavity pressure. For verifying the performance of the NN, 10 sets of data are used in this experiment. Nine sets of data are used to train the NN, and one set of data is used to verify the NN. The experimental results are shown in Fig. 15. The actual cavity pressure Fig. 15 The comparison of the actual cavity pressure and the predicted cavity pressure
775
collected by the pressure probe agrees well with the predicted cavity pressure, and the maximum errors are less than 0.6 MPa, which indicates that this nondestructive cavity pressure measurement method is accurate and effective.
3.4 Advantages of the proposed method Compared with other process monitoring methods, this proposed method for online monitoring of the injection molding process for an injection molding machine is a potentially powerful technology. The following advantages of this method are: 1. Nondestructive. Electrical sensors are installed in the injection molding machine, and not in the mold, for collecting the machine’s physical signals during molding. This method does not destroy the mold. 2. Real time. A multimedia timer technique is employed as a data collection timer. The multimedia timer is activated by a hardware interrupt, with the highest priority. Moreover, the multimedia timer does not send any messages, and there is an independent thread for executing the callback functions. The accuracy of the multimedia timer is 1 ms. 3. Large capacity. A multithreaded technique is designed for collecting and saving signals, and these collected signals are saved in the hard disk and not in the memory. In theory, if the computer hard disk is large enough, then the monitoring process can last indefinitely. 4. Wide open. This method is an open platform, and other nondestructive techniques can be integrated to obtain more information about the injection molding process. For example, the cavity pressure can be measured by the fusion of the data collected from the running machine and the ultrasonic signals. 5. Low cost. The sensors installed in the injection molding machine are inexpensive, traditional electrical sensors: pressure probes, resistive electronic sensors, and K-type thermocouples. In addition, with the rapid development of transducer materials, microprocessors, and signal
776
processing methods, the ultrasonic technology can be executed economically. 6. Health and safety. Ultrasounds are mechanical waves, and the proposed monitoring method is environmental friendly.
4 Conclusions Online monitoring the injection molding process is of great importance for assessing the entire molding process and for ensuring the final product quality. However, until now, nearly all of the monitoring methods destroyed the molds, which was harmful to the appearance and the quality of the final products and reduced the working strength of the molds. In this paper, a nondestructive method is proposed for online monitoring the process of injection molding. Traditional electrical sensors are installed in the injection molding machine, not in the mold, for collecting machine information. As an extension, an ultrasonic monitoring technology is integrated. An ultrasonic probe is placed on the outside surface of the mold for collecting information about the polymer inside the closed mold. During the data collection process, a multimedia timer method, a multithread technique, a stage identification algorithm, and a cavity pressure measurement approach are used. The following conclusions can be drawn: 1. The sampling period of the monitoring method is 1 ms. In addition, the multithread technology can collect signals, save signals in parallel, and store signals on the computer’s hard disk. For a hard disk with a capacity of 250 GB, in theory, this monitoring method can record nine machine signals without interruption for more than 330 days. 2. The area criterion and the angle criterion are suitable for finding the intersections of different stages based on the collected stroke curves. All of the stages of the injection molding process are identified by the developed stage identification algorithm, including the mold closing stage, the injection unit advancing stage, the injection stage, the cooling stage, the plasticization stage, the injection unit returning stage, the mold opening stage, and the ejection stage. The information about these stages, such as the “shock of mold open” phenomenon, can be monitored and visualized. 3. This proposed process monitoring method can be applied to optimize the filling-to-packing switchover point to achieve a smooth pressure change. Moreover, with the help of the injection stroke curve collected by this method, the packing time can be optimized correctly and quickly. 4. Combined with the information collected from the injection molding machine, the ultrasonic technology can be used to quantitatively characterize the key parameters of
Int J Adv Manuf Technol (2014) 72:765–777
the molding process. The ultrasonic reflection coefficient shares similar trends with the cavity pressure. Based on the P-V-T relationship, this nondestructive cavity pressure measurement approach can measure the cavity pressure online. In general, injection molding machines are the most important equipment for the injection molding process, and data collected during operation contains abundant information about the molding process. Hence, the collection and the analysis of information from injection molding machines is a promising method of nondestructively monitoring the injection molding process. Despite the good results achieved in this study, our research will continue with future developments, such as measuring polymer temperatures online and optimizing filling and cooling parameters. Acknowledgements The authors would like to acknowledge financial support from the Science Fund for Creative Research Groups of National Natural Science Foundation of China (no. 51221004), the National Natural Science Foundation Council of China (no. 51105334 and no. 51005151), and the National Key technology R&D Program (no. 2012BAF13B01).
References 1. Osswald TA, Hernández-Ortiz JP (2006) Polymer processing: modeling and simulation. Hanser Gardner Publications, Cincinnati 2. Zhou H, Ying J, Xie X, Liu F, Li D (2010) Nonisothermal crystallization behavior and kinetics of isotactic polypropylene/ethylene– octene blends. Part II: modeling of crystallization kinetics. Polym Test 29(7):915–923. doi:10.1016/j.polymertesting.2010.06.004 3. Meza O, Vega E, Pérez E (2012) Influential factors on the outer lens color in an industrial injection molding process. Int J Adv Manuf Technol 66(1–4):455–460. doi:10.1007/s00170-012-4340-8 4. Guilong W, Guoqun Z, Huiping L, Yanjin G (2010) Analysis of thermal cycling efficiency and optimal design of heating/cooling systems for rapid heat cycle injection molding process. Mater Des 31(7):3426–3441. doi:10.1016/j.matdes.2010.01.042 5. Zhao P, Zhou H, Li Y, Li D (2010) Process parameters optimization of injection molding using a fast strip analysis as a surrogate model. Int J Adv Manuf Technol 49(9–12):949–959. doi:10.1007/s00170009-2435-7 6. Kurt M, Saban Kamber O, Kaynak Y, Atakok G, Girit O (2009) Experimental investigation of plastic injection molding: assessment of the effects of cavity pressure and mold temperature on the quality of the final products. Mater Des 30(8):3217–3224. doi:10.1016/j. matdes.2009.01.004 7. Kurt M, Kaynak Y, Kamber OS, Mutlu B, Bakir B, Koklu U (2010) Influence of molding conditions on the shrinkage and roundness of injection molded parts. Int J Adv Manuf Technol 46(5–8):571–578. doi:10.1007/s00170-009-2149-x 8. Wang J, Xie P, Ding Y, Yang W (2009) On-line testing equipment of P–V–T properties of polymers based on an injection molding machine. Polym Test 28(3):228–234. doi:10.1016/j.polymertesting. 2008.09.003 9. Liu S-J, Su P-C (2009) Novel three-dimensional in-cavity transient temperature measurements in injection molding and fluid-assisted
Int J Adv Manuf Technol (2014) 72:765–777
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
injection molding. Polym Test 28(1):66–74. doi:10.1016/j. polymertesting.2008.10.008 Han X, Yokoi H (2006) Visualization analysis of the filling behavior of melt into microscale V‐grooves during the filling stage of injection molding. Polym Eng Sci 46(11):1590–1597. doi:10.1002/pen.20615 Fathi S, Behravesh A (2008) Real‐time measurement of flow front kinematics using quantitative visualization in injection molding process. Polym Eng Sci 48(3):598–605. doi:10.1002/pen.20945 Özdemir A, Uluer O, Güldaş A (2004) Flow front advancement of molten thermoplastic materials during filling stage of a mold cavity. Polym Test 23(8):957–966. doi:10.1016/j.polymertesting.2004.04.011 Liu S-J, Wu Y-C (2007) Dynamic visualization of cavity-filling process in fluid-assisted injection molding-gas versus water. Polym Test 26(2):232–242. doi:10.1016/j.polymertesting.2006.10.008 Wong HY, Fung KT, Gao F (2008) Development of a transducer for in-line and through cycle monitoring of key process and quality variables in injection molding. Sensors Actuators A Phys 141(2): 712–722. doi:10.1016/j.sna.2007.10.038 Fung KT, Gao F, Chen X (2007) Application of a capacitive transducer for online part weight prediction and fault detection in injection molding. Polym Eng Sci 47(4):347–353. doi:10.1002/pen.20700 Chen X, Chen G, Gao F (2004) Capacitive transducer for in‐mold monitoring of injection molding. Polym Eng Sci 44(8):1571–1578. doi:10.1002/pen.20154 Bur AJ, Thomas CL (1997) Fluorescence monitoring of polymer injection molding: model development. Polym Eng Sci 37(9): 1430–1441. doi:10.1002/pen.11790 Dumitrescu OR, Baker DC, Foster GM, Evans KE (2005) Near infrared spectroscopy for in-line monitoring during injection moulding. Polym Test 24(3):367–375. doi:10.1016/j.polymertesting.2004. 10.003 Ghita OR, Baker DC, Evans KE (2008) An in-line near-infrared process control tool for monitoring the effects of speed, temperature, and polymer colour in injection moulding. Polym Test 27(4):459– 469. doi:10.1016/j.polymertesting.2008.01.010 Zhao C, Gao F, Sun Y (2012) Between‐phase calibration modeling and transition analysis for phase‐based quality interpretation and prediction. AIChE J 59(1):108–119. doi:10.1002/aic.13790 Dontula N, Sukanek P, Devanathan H, Campbell G (1991) An experimental and theoretical investigation of transient melt temperature during injection molding. Polym Eng Sci 31(23):1674–1683. doi:10.1002/pen.760312308 Wang S, Ying J, Chen Z, Cai K (2011) Grey fuzzy PI control for packing pressure during injection molding process. J Mech Sci Technol 25(4):1061–1068. doi:10.1007/s12206-011-0212-7 Dubay R, Pramujati B, Han J, Strohmaier F (2007) An investigation on the application of predictive control for controlling screw position and velocity on an injection molding machine. Polym Eng Sci 47(4): 390–399. doi:10.1002/pen.20720 Zhou H (2013) Computer modeling for injection molding: simulation, optimization, and control. Wiley, Hoboken
777 25. Lionetto F, Maffezzoli A (2008) Polymer characterization by ultrasonic wave propagation. Adv Polym Technol 27(2):63–73. doi:10. 1002/adv.20124 26. Michaeli W, Starke C (2005) Ultrasonic investigations of the thermoplastics injection moulding process. Polym Test 24(2):205–209. doi:10.1016/j.polymertesting.2004.08.009 27. He B, Zhang X, Zhang Q, Fu Q (2008) Real‐time ultrasonic monitoring of the injection‐molding process. J Appl Polym Sci 107(1):94– 101. doi:10.1002/app.27000 28. Abu-Zahra NH (2004) Measuring melt density in polymer extrusion processes using shear ultrasound waves. Int J Adv Manuf Technol 24(9–10):661–666. doi:10.1007/s00170-003-1743-6 29. Ono Y, Jen CK, Cheng CC, Kobayashi M (2005) Real‐time monitoring of injection molding for microfluidic devices using ultrasound. Polym Eng Sci 45(4):606–612. doi:10.1002/pen.20310 30. Ono Y, Whiteside B, Brown E, Kobayashi M, Cheng C-C, Jen C-K, Coates P (2007) Real-time process monitoring of micromoulding using integrated ultrasonic sensors. Trans Inst Meas Control 29(5): 383–401. doi:10.1177/0142331207080153 31. Fronimidis D, Edge K, Stelson K, Mileham A, Sokola M (2004) Use of non-invasive measurement methods to aid the control of part quality in the injection moulding process. Proc Inst Mech Eng B J Eng 218(9):1195–1201. doi:10.1243/0954405041897158 32. Zhao P, Fu J, Cui S (2011) Non‐destructive characterisation of polymers during injection moulding with ultrasonic attenuation measurement. Mater Res Innov 15(s1):s311–s314. doi:10.1179/ 143307511×12858957674355 33. Gao F, Patterson W, Kamal M (1996) Cavity pressure control during the cooling stage in thermoplastic injection molding. Polym Eng Sci 36(19):2467–2476. doi:10.1002/pen.10644 34. Zhou H, Li D (2005) Computer evaluation of weld lines in injectionmolded parts. J Reinf Plast Compos 24(3):315–322. doi:10.1177/ 0731684405043558 35. Shi H, Xie S, Wang X (2013) A warpage optimization method for injection molding using artificial neural network with parametric sampling evaluation strategy. Int J Adv Manuf Technol 65(1–4): 343–353. doi:10.1007/s00170-012-4173-5 36. Huang MS (2007) Cavity pressure based grey prediction of the filling-to-packing switchover point for injection molding. J Mater Process Technol 183(2):419–424. doi:10.1016/j.jmatprotec.2006.10. 037 37. Kazmer DO, Velusamy S, Westerdale S, Johnston S, Gao RX (2010) A comparison of seven filling to packing switchover methods for injection molding. Polym Eng Sci 50(10):2031–2043. doi:10.1002/ pen.21731 38. Wang J, Peng J, Yang W (2011) Filling-to-packing switchover mode based on cavity temperature for injection molding. Polym-Plast Technol 50(12):1273–1280. doi:10.1080/03602559.2011.578292 39. Pantani R, De Santis F, Brucato V, Titomanlio G (2004) Analysis of gate freeze‐off time in injection molding. Polym Eng Sci 44(1):1–17. doi:10.1002/pen.20000
