New Ductile Cast Iron Digital Grades for Automotive Components

Several contributors

Using data with monitored automotive brake anchors for more than one year and expert analysis of industrial manufacturing, researchers have developed a new, high yield-strength material grade of iron. The research project demonstrated that the well-known EN GJS-550 high-strength standard material grade with yield strength exceeding 370MPa—commonly used for safety components—can be replaced with the new HS420/250 grade. This new grade exhibits a yield strength higher than 420 MPa and elongation higher than 5%, with the resulting hardness maintained below 250HB (Brinell hardness).

This 13.5% increase in yield strength opens new design opportunities for lightweight cast iron components.

The development of light metal components based on high-strength steels and aluminum has increased significantly in recent decades, while little has been invested in the development of high-strength cast iron. The general high yield-strength grade of cast iron for safety automotive casting components is EN GJS-550; yield strength exceeding 370 MPa according to specification compared to yield strength of up to 900 MPa can be achieved with steel materials.

In the development of highstrength steels, the strongest yield strength is attributed to the grain-size refinement associated to controlled thermo-deformation rolling and interaction with precipitation of microalloying elements such as Nb, Ti, and/or V. This mechanism cannot be transferred to cast iron, where morphology of graphite and ferritic/pearlitic ratio are the main microstructural features, and the process does not present any deformation step. However, the use of selective additions for promoting pearlite formation in cast irons material structure has become common practice for many decades, with emphasis on elements such as copper, manganese, arsenic, and tin. The higher the pearlite content, the higher the yield strength and maximum load, while resulting elongation is generally lower. Cu and Mn within the material composition act on the eutectoid transformation, which has led to several potentially contradictory explanations. It has been found that small amounts of Cu can counteract the effect of low-level addition of Mn. Another study on the effect of small increments in copper contents from 0.11 up to 1.0 wt. % led to a clearer understanding of its effect on the eutectoid transformation. Within industrial practice, different combinations of these alloying elements are observed in ductile cast iron when EN GJS-550 quality is manufactured and sometimes a yield strength higher than 420 MPa is observed.

Conversely, in recent years, the traditional manufacturing industry is challenged worldwide with the growth and advancement in digital technologies that facilitate the integration of interconnected intelligent components inside operations of the shopfloor. Digital twin-based, sustainable intelligent manufacturing is an emerging technology that can grasp the state of intelligent manufacturing systems in real time and predict system failures.

The use of digitalization technologies has enabled virtual product and process planning. The resulting large amounts of data are processed, analyzed, and evaluated by simulation and optimization tools to make them available for planning in real time. One of these simulation-based planning and optimization concepts with great potential in many industrial fields is the model predictive control system, more commonly known as digital twin, which is the virtual and computerized counterpart of a physical system. The digital twin in its origin describes mirroring a product, while the state-of-the-art technology allows processes to be subjects of virtual space reproduction. The complexity of manufacturing processes usually requires a modelling of certain aspects within these processes to understand and optimize them. These physical models predict relevant output on the basis of relevant inputs. Meanwhile, other approaches are based on black boxes that use big data techniques for output prediction. The common targets of this kind of technology is to increase competitiveness, productivity, and efficiency, among others. 

In this project, the manufacturing digital twin already in operation in a cast iron foundry was used as a starting point for the development of the new material grades. Manufacturing data of more than 170 automotive brake anchor castings were collected, covering both the conventional and new high yield-strength grade (> 420 MPa). A model predictive control (MPC) was created based on artificial intelligence tools. The algorithm was integrated into the foundry’s digital twin that is able to perform online predictions concerning the yield strength. Automotive brake system castings were manufactured following the predictions of the MPC, fulfilling the high yield-strength mechanical properties. 

The highest yield-strength cast iron grade currently used to manufacture safety automotive brake cast anchors is according to EN GJS 550 quality. There is no direct requirement for the ferrite/pearlite ratio within the material microstructure. However, the mechanical properties and hardness specifications shown in Table 1 must be fulfilled within the cast anchors products. The specification target for the new high yield-strength grade (HS420/250) is also shown in Table 1. The objective was to increase the yield strength and maximum load, while the elongation and hardness values are maintained within the standard grade EN GJS 550.

For each selected batch, the melt quality in the pouring unit was analyzed using a Thermolan thermal analysis system developed by Azterlan. Each batch of approximately 500 anchors was manufactured with the analyzed metal until a new ladle was delivered inside the production pouring unit. All critical parameters measured in liquid-solid transformations and in solid state (eutectoid) transformations are shown in Figures 1a and b. All necessary data were automatically stored in the AAPICO foundry database. Recalescence, ∆T is calculated as the difference between Tmax and Tmin, in both liquid-solid transformation (∆Ts) and in eutectoid transformation (∆Teu). Vtrans is defined as the cooling rates in the middle of the eutectoid transformation, being negative for the ferritic alloy and positive associated to the recalescence in the pearlitic alloy. 

Figure 1b compares the eutectoid cooling curves of a ferritic and a pearlitic cast iron alloy. The eutectoid transformation of the pearlite alloy occurs at lower temperatures, and it presents a clear recalescence, ∆Teu, in comparison to the ferritic alloy.

AAPICO foundry has a data acquisition system (DataPRO) able to gather all key manufacturing process parameters, collecting more than 300,000 data points per day for the last 10 years. The melt quality assessment is performed just after delivering the new liquid metal into the production pouring unit. After casting, several components are extracted from the manufacturing line and used for the quality control assessment.

A manufacturing digital twin (OLIMPO) has been in operation for the last five years, significantly reducing the quality defects and improving the overall equipment effectiveness. An example of the digital twin screenshots is shown in Figure 2. Based on the production data from DataPRO and applying model predictive controls, the digital twin provides predictions that allow the operator to know the conditional state of the metallurgical quality without having to make costly final verifications.

For the development of a new high yield-strength cast iron grade, a revision of the data collection was performed. The main parameters collected were the following: (1) Batch code of each reference production, identification, and traceability with records of each nodulization treatment; (2) Identification and recording of the main metallurgical parameters of each treatment, by means of thermal analysis, minimum eutectic temperature, self-feeding factor, nodularity index, nodule counts, eutectoid transformation, chemical composition, pouring time, applied inoculation, and shakeout temperature of the casting parts.

Two different commercial automotive safety brake anchor castings with a weight between 1.0−1.3 kg were selected for this study. In total, 173 batches of the two anchors were analyzed. Approximately 500 castings were manufactured within each batch.

A detailed control procedure was developed for the monitoring and analysis of each production batch. Two molds were identified for each manufacturing batch, and the casting corresponding to the same cavity was removed from the end of the production cooling line. The castings were mechanically tested, with all results and data collection stored within a new database.

To evaluate the most influential variables, Pearson’s correlation was applied, and the most significant parameters were obtained. Next, several self-learning algorithms (rules, decision trees, Bayesian networks, and neural networks) were tested, using non-discretized and discretized data.

After data analysis, the predictive model with the highest accuracy was selected and incorporated in the smart module and integrated into AAPICO’s manufacturing digital twin following the schema shown in Figure 3.

Figure 4 compares an example of the stress-strain curve of the current EN GJS-550 grade and the new high yield-strength grade, HS420/250. Figure 5 shows the mechanical properties (yield strength versus elongation) from the 173 data points collected for this study. A high variation is observed in both yield strength (varying between 370 and 480 MPa) and in elongation (varying from 5%-11 %). If the yield strength is higher than 420 MPa, it fulfills the new high yield-strength grade condition shown in Table 1 and is marked in orange color. Thirtytwo data points are in the range of this new high yield strength grade (HS420/250).

First, to evaluate each variable influence in mechanical properties, concretely in yield strength, Pearson’s correlation was applied to all collected manufacturing process parameters. Approximately 100 variables were studied. The most significant variables for obtaining high mechanical properties are shown by importance order in Figure 6. These variables are alloying elements such as Cu and Mg plus eutectoid transformation thermal analysis parameters. To evaluate the effect of each individual parameter, the yield strength is plotted versus each individual significant parameter in Figures 7−10.

Because copper is a pearlitizer element, a higher pearlite content was expected as copper content increased. The higher the pearlite content, the higher the yield strength was expected. However, a clear data dispersion in Figure 7 concludes that additional parameters influence yield strength. In addition, the variation of pearlite amount produces significant changes in the eutectoid cooling curves. When pearlite content increases, solid state transformations occur at lower temperatures and eutectoid recalescence, ∆Teu, clearly increases. A higher latent heat is expected when eutectoid cementite is formed in the pearlite instead of full ferritic microstructure. A gradual reduction in VTran is also observed associated with larger pearlite amounts in cast iron. Conversely, the magnesium content is related to the nodulization potential of the melt, which also increases the carbides formation tendency in the pearlite.

Once the variables were selected, two datasets were prepared: one with discretized and another one with non-discretized data for AI analysis. A comparative study with several well-known supervised machine learning techniques namely, Bayesian Network, Decision Tree, KNearestNeighbor, Artificial-Neural-Network, and Support Vector Machine, was conducted. To evaluate the performance of each machine learning technique, confusion matrix and algorithm model accuracy were used. The best model obtained was a Bayesian network model, whose data has been previously discretized by Kononenko methodology.

Table 2 shows the confusion matrix obtained with the Bayesian network model for the two selected classes: high yield strength (HS420/250) and conventional grade (EN GJS550). Accuracy, error rate, sensitivity and specificity were calculated and summarized in Table 3. The obtained accuracy with the dataset consisting of 173 analyzed batches by crossvalidation of 10 folds is higher than 90 %. A big population of data in Figure 5 lies in the conventional grade range between 400 and 420 MPa. Using the lessons learned in this study, with minor adjustments in the copper, magnesium, and other elements that affect the eutectoid transformations, it would be possible to achieve a slightly higher yield strength and become part of the new material grade.

Self-learning tools also can be applied periodically. As soon as the amount of data of the new material is about the same as the amount of data of the conventional material, the accuracy of the model predictive control, based in the Bayesian algorithm, will increment significantly.

Figure 11 shows the main relationship of the Bayesian network model where the key parameters are related. Appropriate combinations must be assured in the manufacturing of new high yield-strength grade material. Thus, optimized material is predicted in the following conditions:

• 77.3% of the data presents a Cu content higher than 0.28 wt.%.

• 94.1% of the data presents a Mg content higher than 0.036 wt.%.

• 94.12% of the data presents a recalescence, ∆Teu higher than 1.71C/3.08F.

• 97.1% of the data presents a VTran higher than 0.034C (0.061F)/s.

In the manufacturing of a new batch, all these key parameters were controlled by the smart module described in Figure 3 and integrated in the foundry manufacturing digital twin. If the system predicted unfavorable results, corrective actions based in ad-hoc actuation protocols were performed to modify the defined key parameters before liquid iron melt pouring and casting manufacturing processes.

Several trials were performed based on the defined chemical target compositions and manufacturing conditions to produce conventional brake anchors with conventional material EN GJS-550 and lightweight anchors with both old and new materials.

Some examples of the chemical composition and microstructure observed in both conventional and new HS420/250 are shown in Table 4 and Figure 12. The most influencing parameters of pearlite content measured within the anchor casting plus the mechanical properties are also summarized in Table 5.