Industry 4.0 – Building smart digital bridges

Industry 4.0 – Building smart digital bridges

Modern production facilities are loaded with sensors, systems and smart controls, all providing tons of data on the process and results. The spread of Plant Floor Data Collection (PFDC), Process Condition Monitoring (PCM) and Manufacturing Execution Systems (MES) in modern manufacturing facilities keeps growing and with that, digitizing Manufacturing Processes continues to spread just as fast as it does in our society. As soon as more data becomes available, companies are faced with the common challenges, which unfortunately are not always reflected when designing the solutions. How can we benefit from the available data? How can we benefit from combining data? In other words, how can data be transformed into information.

For a manufacturer of aluminum engine components, we did a design study and coaching of the implementation for the casting department based on the outcome of the ReVista project and own experiences in the industry. The Operational Excellence team in the plant had prepared priority list based on initial financial analyses. The projects followed a structured approach, commonly known as Plan – Do – Check – Act:

  1. PLAN – Design Thinking sessions to determine pain points and possible solutions and technical design of the solution.
  2. DO – Implement the solutions.
  3. CHECK – Monitor the results.
  4. ACT – Make the necessary corrections.
  5. Continue with step 1!

The first priority on the list was the filling of the holding furnaces of the casting machines. These holding furnaces have submerged heating elements to maintain the required temperature for the molten aluminum and the available data showed that there is a clear correlation between the filling level of the holding furnace and the power consumption of the heating elements. Further analyzes also showed a correlation between the life time of these elements and the filling level, see also the ReVista project publication section 4.5. Material quality was also impacted by the filling level, since both cold draft shock and time of exposure to air of the surface of the molten material had impact on oxidation and contamination of the material. And of course the casting machine would run out of material when the holding furnace wasn’t filled in time.

Facts to be considered for the solution:

  • Forklift drivers filled ladles with 1.000 kg molten aluminum and after material treatment cycled through all casting machines to fill the holding furnaces continuously.
  • Each holding furnace could hold maximum 800 kg.
  • Each produced part would consume between 8 kg and 80 kg of aluminum, depending on the product being produced.
  • The aluminum cools down in the ladle during the process and transport because the ladles are not equipped with heating elements.
  • All parts can be produced on all machines, and the required temperature of material depends on the product, ranging from 700C to 760C. Different types of products require different processes for material treatment, there are 7 groups of products with different processes. The mix varies based on demand.
  • The holding furnaces need to be filled at a material temperature range of -10C and +20C which is set in the recipe for the part being produced. The heating elements of the holding furnaces can compensate the -10C without causing a stoppage or quality issues. When the temperate of material gets out of range, the machine has to be stopped to avoid quality issues.
  • All involved equipment is equipped with sensors and network connections but all are used is independent systems and controls.

Tests by the team had shown that the optimal filling level of the holding furnaces is between 50% and 80%. When filled over 80%, the heating elements consume more electricity to maintain the temperature and the margin for temperature fluctuations which can be compensated reduces significantly. When filled below 50%, the time the material is exposed to the air inside the holding furnace causes increasing oxidation and contamination of the material. Below 30% filling, the submerged heating elements are oversized for the available amount of aluminum and the temperature starts to vary more than desired. In addition, the negative effect of cold draft shock on material, furnace and heating elements starts to increase rapidly. When 10% or less of the holding furnace is filled, the heating elements get directly exposed to the air in the holding furnace, causing increased wearing and shortening the life time of the elements. With this, we created a simple priority schedule for filling the holding furnaces:

  • Optimal: 50% – 80% filling, plan for refill before reaching 50%.
  • Avoidable: 80%+ filling, set non-critical marker for learning scenario to avoid this status.
  • Undesirable: 30% – 50% filling, schedule for urgent filling and set urgent marker for learning scenario to avoid this status.
  • Critical: 10% – 30% filling, schedule for critical filling and set critical marker for learning scenario to avoid this status.
  • Emergency: Below 10% filling, shutdown machine and schedule for filling after release by the operator, and set critical marker for learning scenario to avoid this status.

First phase

We found that although there was no measurement of the actual filling level of the holding furnaces, we could estimate the level by using the shot count and the part ID provided the machine, and the weight of the produced part which was available in the ERP and MES. Constantly updating the estimated filling level which each shot and with each filling, we had a moving indicator of how much material was in the holding furnace with one major challenge. There was no way in which we could measure how much material the forklift driver would actually fill in the furnaces which each loading.

For this, we decided on a multi-layer approach to continuously improve the accuracy of data and the filling process. First step was training the forklift drivers on always aiming for the 80% filling marker and explaining WHY this was so important. We also implemented a plausibility check on the bookings of filling by the forklift driver compared to the estimated and required filling of holding furnaces. Since the operator would open the holding furnace every two hours for cleaning during production, the control panel for the furnace was extended with a validation screen for content of the furnace. The operators could see markings in steps of 10% inside the furnace so this provided the system with feedback when the deviation would exceed 10%. We activated this validation automatically every time the holding furnace was opened, so also the forklift drivers and supervisors had the opportunity to give feedback.

Testing this data collection for two weeks showed a rapid improvement of data accuracy and the team was confident enough to start showing forklift drivers filling priority based on the available data. We designed a straightforward priority screen, showing the next 4 machines that would need filling with their priority status. The AND-ON for the holding furnace was also modified so it would show the priority status of the filling. As KPI we decided to measure the Critical and Emergency status of holding furnaces, and the energy consumption of the heating elements.

Already in the first week, the status Emergency was completely eliminated and we found that the forklift drivers did a very good job with the available information. What this first simple solution didn’t help the team with was preventing loading the furnaces with too low temperature material, time to drive between the holding furnaces, prediction of consumption, etc. We knew we had to address these issues and we also knew that this first implementation wouldn’t solve everything at once. The pilot implementation did however show very clearly that we could eliminate the most costly pain point: having to stop a machine because it had run out of molten material!

Second phase

Temperature development of the material in the ladle is far from linear and equipping all ladles with submerged temperature sensors rather expensive and complicated due to the material preparation which is done in the ladle before filling the furnaces. One of the engineers found a very practical solution for this important challenge. A calibrated and isolated temperature sensor placed at exactly 8mm from the inner lining of the bottom of the ladle provided temperature data with a 0.5% accuracy. With this solution we had a temperature sensor which was accurate enough and at the same time fully protected from the aggressive material preparation and the molten aluminum itself, which can be just as aggressive.

The plant had over 2 years of RFID based data on the actual filling of the holding furnaces and from this we could extract the driving times in all kinds of combinations. Time between the melting furnace and the material preparation was stable in the various combinations of melting furnaces (4) and the material preparation stations (3). The required time between the material preparation stations and the first holding furnace that would be loaded was also very stable, although there was an odd difference between the first loading being on the right side or the left side of the factory. After the first filling, the variations were very high and we were not able to find a logical explanation by analyzing the data.


The very logical explanation was found quickly by spending time in the plant and observing the forklifts and the loading process. The ladle had a filling opening on the right side, meaning that the forklift had to have the holding furnace on the right side. Depending on the location of the previous holding furnace, the forklift either had the next holding furnace already at the right side or had to make a turn first before maneuvering to the next furnace. The time the forklift driver needed for this turn depended on available place and the biggest influence was whether or not there was another forklift filling a furnace of the nearby or opposite machines. With liquid aluminum in the ladle, the forklift drivers made sure this maneuvering was done with care! Now we also understood why the first filling at the right bank of casting machines commenced so much faster than when the first filling was at the left side, although there was hardly any difference in distance.

We learned more important factors and variables from these GEMBA sessions. The holding furnaces can only be opened outside of the pressure cycle of the casting machine, which is between 1 and 8 minutes long, depending on the recipe of the product. Having the forklift wait in front of the machine for the cycle to end is basically a downtime of the forklift and has negative impact on the temperature of material in the ladle. We had seen that the accuracy of predicting the amount of aluminum in the furnace and the next needed filling for machines producing large parts was much higher than for the smaller parts. The large components which could consume up to 80 kg of aluminum per product had the longest pressure cycles and lowest amount of stoppages during the day. The smaller parts had shorter pressure cycles and much more stoppages during the day, which makes sense when all steps of the process are executed at up to 10 times higher frequency compared to the largest parts.

We also observed casting operators declining filling of the holding furnace before they would clean the furnace. Observing and asking questions gave an obvious explanation for this, which we would have never picked up from the comfortable meeting rooms. The less material in the holding furnace, the faster the operator can clean the furnace of contaminated material and residue. When the furnace is filled for 50% or more, which happens to be the optimal filling level during production, it is almost impossible for the operator to perform a proper cleaning.

Drawing board session!

We had the required temperature for the holding furnaces from the recipes for the products. Not only could we extract this from the machine itself and the MES, the planning component of the MES even provided temperature requirements for upcoming planned production. We had live data on the actual material temperature in the ladle and found influences we had not reflected yet. The material cooled down significantly faster when there was less material in the ladle, and the material also cooled down faster when the ladle was in tilted position to fill the holding furnace and this influence increased when there was less material in the ladle. Depending on the type of material preparation, the starting temperature was lower when the process required more time. The driving pattern of the forklift influenced the remaining temperature.

The MES triggered the cleaning process of holding furnaces as controlled quality process, so we could read the schedule for this. The PFDC system provided live data about the status of the machines and stoppages, and even a “stoppage likelihood” indicator which was calculated by an algorithm that reflected current state and patterns from previous production of the product. A relatively simple additional modeling in the MES to reflect the accurate material preparation process allowed us to not only reflect the required time for this process and the expected temperature after the process, but also to make sure that a ladle filled with prepared material for a certain product group would not be assigned to fill a holding furnace that required differently prepared material.

Putting it into action

The Plant Systems team, responsible for the MES, PFDC and PCM, modelled all variables in their scheduling system, using the same core algorithms they are using for their production planning system. The end result of this second stage of development is that the forklift drivers see a schedule on their panels to fill the holding furnaces, which is updated in real-time. The system behind this schedule is reflecting production demand and schedules, technical parameters, simplified temperature predictions and the predicted time frame in which the furnace can be opened for filling. After 4 weeks of operational usage and continuous improvement of data and logic, the financial results are very positive:

  • Waiting time of forklifts before being able to open the holding furnace is completely eliminated.
  • Machine stoppages due to lack of material are completely eliminated.
  • Electricity consumption of the heating elements is reduced by 8% by maintaining the “sweet spot” of holding furnace load.
  • Temperature related quality issues caused by loading errors are completely eliminated.
  • Accidental loading of holding furnaces with wrong material is completely eliminated.
  • As a really important bonus, the working relationship between the forklift drivers and casting operators has improved significantly!

Next steps

The team is currently evaluating what benefits and challenges would arise when the ladles are modified so the forklifts can fill the holding furnace from the left and the right side. Three different scenarios are currently evaluated and will be tested as prototypes soon. Scenario 1 is to modify the mounting of some ladles so they can be used for left sided filling. Although this is the fastest and cheapest solution, it would create a rigid structure with either left or right filling but it could however eliminate the complex maneuvering during rounds. Scenario 2 foresees a more complicated modification in which all ladles can be mounted to the forklift either right or left sided. The costs are higher because the mounting frame includes safety features and data connections. Scenario 3 provides a very flexible solution and makes one wonder why this wasn’t done in the first place. This scenario foresees that all ladles will have a filling lead at both sides, enabling the forklift driver to always select the optimal side to fill the furnaces. This scenario, as efficient and simple as it might appear, does however require that all ladles will have to be replaced.

Predicting the temperature is now done based on the scheduled route for filling the furnaces in a simplified model. Real-time data from the temperature sensors show that these predictions are good for the beginning of the filling tour but not good enough towards the end of the tour. The team wants to add AI to predict the temperature development much better and feed these predictions to the scheduling system for all possible combinations, allowing the system to even reflect the predicted temperature development while picking the optimal scenario. Better temperature prediction would also allow the system to schedule filling tours with decreasing temperature requirements, for example two filling where the temperature had to be in a range of 740C – 760C and a filling where the temperature has to between 720C and 740C.

In the current setting, the filling schedule is the result of the approved production planning and real-time data from the production environment. The production planning team is very interested in including the holding furnace area in their capacity planning modules. This would allow WHAT-IF and decision making about needing 3 or 4 forklift drivers for the casting production schedules. At the moment they already use similar evaluations for the melting furnaces and this addition would allow the team to analyze more details on the impact of various scenarios.

The learning markers are now evaluated manually by the supervisors with support of the OPEX team on a daily basis. The team wants to add more learning markers to the system, for example for temperature deviations, and in the future add an AI component to analyze the learning markers to constantly improve the scheduling algorithms.

Related topic: Industry 4.0 – Now what?

2 thoughts on “Industry 4.0 – Building smart digital bridges”

  1. Pingback: USE CASE Industrial ML and Cloud in Manufacturing - AWS re:Invent

  2. Pingback: USE CASE Industrielles ML und Cloud in der Fertigung - AWS re:Invent

Comments are closed.