Merging the Worlds of Machine and Simulation
Adjusting an injection molding machine until the optimum operating point is reached requires lots of time and expertise. Researchers at Bosch are aiming to reduce the time and effort required – from the production launch until the desired series-production quality is achieved – by simulating the entire process on a computer. Doing so will provide them with superior start parameters for the production launch. A self-optimizing closed-loop control in the injection molding machine ensures that the plastic parts are of good quality.
Plastics display strange behavior. Compared to many other materials – including metals – plastics behave in a far more complicated manner. And that’s true, for example, when a plastic cools off, because it then contracts differently, and in different directions, in what experts refer to as anisotropic material behavior. This complex behavior is a source of problems because it makes it difficult to produce high-precision plastic components at the quality levels that are required by Bosch.
Injection molding techniques can be used to bring the obstinate material into practically any form desired. Because of this attractive feature, the technique is widely used for intake pipes, valve covers, power strips and housings for electronic devices. The anisotropic material characteristics unfortunately make it difficult to predict if a mold can produce a component with the precision required. This makes tool production and the machine adjustment process very time-consuming and expensive: The breaking-in phase (sampling) can take several weeks. That’s why Bosch researchers are aiming to increase precision and substantially reduce set-up times by simulating the entire system.
The injection molding technique is easy to explain. It involves pouring thermoplastic polymer granules (which become plastic when heated) into a funnel. A screw transports the material through a heated cylinder, where it is heated to around 300 °C. This causes it to melt and turns it into a moldable, highly viscous mass similar to honey. The second step in the process involves injecting the material at high pressures (up to 2,000 bars) into the mold, where it is given its shape. After the plastic has cooled and solidified, the mold opens and the finished injection-molded part falls onto the production line or is extracted by a robot arm.
Depending on the component, the entire production cycle lasts for about 20 to 90 seconds, and the cooling process takes up more than half of this time. The cooling process therefore plays an important role in production and clearly demonstrates how the various parameters interact. If the designers make the plastic somewhat thicker in order to strengthen it, for example, the greater mass causes it to cool at a slower rate.
But the output rate will then decrease, which pushes up unit costs. A rapid cooling process can cause the component to deform in sometimes unpredictable ways, however, resulting in sub-standard quality.
In determining the optimum process parameters for a conventional injection molding machine, several weeks of statistical test modeling are usually needed. The tests take 10 to 15 different machine settings into account – e.g., mold temperature, pressure, polymer melt temperature and injection speed – and these are then statistically varied. The results of these tests are used to gradually attain the optimum machine setting.
To accelerate this process, the Bosch researchers now want to use computer simulations. To ensure that all factors influencing the injection molding process are taken into account, the overall system is divided into four areas: material properties, machine characteristics, the mold itself and external influences. The aim is to achieve the highest possible quality for molded parts by optimizing process parameters at first in computer simulations and then in real-life tests.
By 2010, the researchers at Bosch plan to have a virtual recreation of a complete injection molding machine. This will allow them to determine the machine’s optimum operating variables in advance and even predict quality. Although the simulations naturally do not make the tests superfluous, the time needed for them would be substantially reduced.
The Bosch researchers recently succeeded in completing an important step toward achieving their goal by using statistical observations of the process parameters to offset fluctuations and drifts. To do so, the researchers used linear regressions between the controlled variables and quality characteristics as well as neural networks. The latter are particularly useful when dealing with non-linear relationships, such as the anisotropic warpage characteristics of plastic.
A special feature of the system is that the control software automatically selects the best method (linear regression or neural network) and also decides if it must be recalibrated or if the neural network has to be recalculated. As a result of the control mechanism’s autonomous behavior, the development of a self-optimizing injection molding machine is now within reach.
As part of a project that was organized in cooperation with industrial partners and research institutes, Bosch recently succeeded in demonstrating the capabilities of control models based on neural networks. The project called for evaluating the millions of setting combinations for an injection molding machine’s operating point. Guaranteeing the very best quality would normally require that the process be studied with the help of comprehensive statistical test records.
By training a neural network, though, it was possible to reduce the enormous number of possible settings to approximately 370 combinations. And following a further process of fine tuning, the number of different operating points was even slashed to 20. As a result, considerable savings in the amount of time and effort required to optimally set the machine were realized. Bosch is researching injection molding machines on two different levels: On the one hand, the company’s researchers are using model-based process analysis, process control and process optimization measures to study the real-life machines and their environments.
This involves measuring phenomenological values including temperature, pressure and flow rate, which are then incorporated into the control loops. The other level involves virtual machines – computer simulations that are based on the basic physical principles of materials physics. And although the non-linearities that appear during this process still pose a major challenge for the researchers – especially when material properties are being simulated – Bosch’s special expertise is clearly demonstrated in its ability to interconnect the worlds of hardware and simulation.
Injection molding techniques can be used to bring the obstinate material into practically any form desired. Because of this attractive feature, the technique is widely used for intake pipes, valve covers, power strips and housings for electronic devices. The anisotropic material characteristics unfortunately make it difficult to predict if a mold can produce a component with the precision required. This makes tool production and the machine adjustment process very time-consuming and expensive: The breaking-in phase (sampling) can take several weeks. That’s why Bosch researchers are aiming to increase precision and substantially reduce set-up times by simulating the entire system.
The injection molding technique is easy to explain. It involves pouring thermoplastic polymer granules (which become plastic when heated) into a funnel. A screw transports the material through a heated cylinder, where it is heated to around 300 °C. This causes it to melt and turns it into a moldable, highly viscous mass similar to honey. The second step in the process involves injecting the material at high pressures (up to 2,000 bars) into the mold, where it is given its shape. After the plastic has cooled and solidified, the mold opens and the finished injection-molded part falls onto the production line or is extracted by a robot arm.
Depending on the component, the entire production cycle lasts for about 20 to 90 seconds, and the cooling process takes up more than half of this time. The cooling process therefore plays an important role in production and clearly demonstrates how the various parameters interact. If the designers make the plastic somewhat thicker in order to strengthen it, for example, the greater mass causes it to cool at a slower rate.
But the output rate will then decrease, which pushes up unit costs. A rapid cooling process can cause the component to deform in sometimes unpredictable ways, however, resulting in sub-standard quality.
In determining the optimum process parameters for a conventional injection molding machine, several weeks of statistical test modeling are usually needed. The tests take 10 to 15 different machine settings into account – e.g., mold temperature, pressure, polymer melt temperature and injection speed – and these are then statistically varied. The results of these tests are used to gradually attain the optimum machine setting.
To accelerate this process, the Bosch researchers now want to use computer simulations. To ensure that all factors influencing the injection molding process are taken into account, the overall system is divided into four areas: material properties, machine characteristics, the mold itself and external influences. The aim is to achieve the highest possible quality for molded parts by optimizing process parameters at first in computer simulations and then in real-life tests.
By 2010, the researchers at Bosch plan to have a virtual recreation of a complete injection molding machine. This will allow them to determine the machine’s optimum operating variables in advance and even predict quality. Although the simulations naturally do not make the tests superfluous, the time needed for them would be substantially reduced.
The Bosch researchers recently succeeded in completing an important step toward achieving their goal by using statistical observations of the process parameters to offset fluctuations and drifts. To do so, the researchers used linear regressions between the controlled variables and quality characteristics as well as neural networks. The latter are particularly useful when dealing with non-linear relationships, such as the anisotropic warpage characteristics of plastic.
A special feature of the system is that the control software automatically selects the best method (linear regression or neural network) and also decides if it must be recalibrated or if the neural network has to be recalculated. As a result of the control mechanism’s autonomous behavior, the development of a self-optimizing injection molding machine is now within reach.
As part of a project that was organized in cooperation with industrial partners and research institutes, Bosch recently succeeded in demonstrating the capabilities of control models based on neural networks. The project called for evaluating the millions of setting combinations for an injection molding machine’s operating point. Guaranteeing the very best quality would normally require that the process be studied with the help of comprehensive statistical test records.
By training a neural network, though, it was possible to reduce the enormous number of possible settings to approximately 370 combinations. And following a further process of fine tuning, the number of different operating points was even slashed to 20. As a result, considerable savings in the amount of time and effort required to optimally set the machine were realized. Bosch is researching injection molding machines on two different levels: On the one hand, the company’s researchers are using model-based process analysis, process control and process optimization measures to study the real-life machines and their environments.
This involves measuring phenomenological values including temperature, pressure and flow rate, which are then incorporated into the control loops. The other level involves virtual machines – computer simulations that are based on the basic physical principles of materials physics. And although the non-linearities that appear during this process still pose a major challenge for the researchers – especially when material properties are being simulated – Bosch’s special expertise is clearly demonstrated in its ability to interconnect the worlds of hardware and simulation.
