Humachine dynamics for the four cases’ applications
| Case . | Requirement . | Humachine dynamics classification . |
|---|---|---|
| Asphalt | Process setup optimisation | 1a is PDCE when it refers to an existing product for which the current mix recipe is not satisfactory and it is of a DME when about a new product |
| Type implemented: Type 1 (in both cases) | ||
| Analysis: in the current implementation of the mix recipe optimisation several prediction tools are applied but no direct actuation of the automatic decision is ever foreseen, therefore the implemented humachine is always Type 1, while for DME target (new product design) a Type 3 is the desirable option, and for PDCE target (existing product optimisation) the implemented humachine is in line with the desirable option (Type 1) | ||
| 1c is ART since it is deterministically executed every time material is dispatched for defining the target loading temperature | ||
| Type implemented: Type 1 | ||
| Analysis: Type 1 mismatches the desirable option (Type 2) because of the incapability to precisely predict temperature at paving site | ||
| Process r/t control | 1b is DME since technology aims at making diagnoses | |
| Type implemented: Type 1 | ||
| Analysis: Type 1 mismatches the desirable option for a DME target (Type 3) for the automatic reconfiguration. Indeed, ML technologies provide suggestions only, while human validates anomalies and determines mitigation actions | ||
| Steel | Planning | Planning is activated by a human operator (2b) to ensure optimal scrap usage. It is a DME task, since planning tool proposes a solution, and human intervention occurs only if the proposed solution is deemed suboptimal. Moreover, scenario exploration is a typical feature of planning tools, as simulations allowing exploration of alternative scenarios. However, the application of the final plan might be the result of human-MAPE-K cooperation even if systematically human supervised |
| Type implemented: Type 3 | ||
| Analysis: in this case, the implemented type is in line with the DME desirable option | ||
| Process setup optimisation | A simulation tool for the EAF recipe (2a) is integrated into the automation system to identify root causes of production non-conformities. Aligning with the need for automated diagnosis, it is an ART-type application | |
| Type implemented: Type 1 | ||
| Analysis: Type 1 does not match the desired configuration for ART target, which would require Type 2 | ||
| 2c is a PDCE when it refers to re-designing recipes, and a DME when it refers to a new design | ||
| Type implemented: Type 1 (in both cases) | ||
| Analysis: in the current implementation of the design/re-design of a recipe making several prediction tools are applied, but no direct actuation of the automatic decision is foreseen; therefore, the humachine type for the DME target is not the desired one (Type 3), whereas it corresponds to the desirable option and for a PDCE target (Type 1) | ||
| Pharma | Process setup optimisation | The process settling optimisation (3a) can be classified as a PDCE task since the process might be standardised once enough data is collected in all possible working points |
| Type implemented: Type 1 | ||
| Analysis: the implemented type is coherent with the desired one for a PDCE target. | ||
| Process r/t control | An MPC acts continuously on the basis of an incapsulated predictive ML model to regulate a chemical reaction (3b, target classification ART). The controller freezes if the control trim exceeds a tolerance level and, in this case, human intervention is required for correction. In the nominal conditions, the reference generated by the MPC controller is actuated in an unsupervised way. ML model is also applied to automatically define the controller initial settings | |
| Type implemented: Type 2 | ||
| Analysis: the implemented type is coherent with the desired one for an ART target. | ||
| The ML model is also used to diagnose possible OCT misplacements (3c, target classification PDCE). In case of anomaly detection, the OCT positioning remains human actuated | ||
| Type implemented: Type 1 | ||
| Analysis: the implemented type is coherent with the desired one for a PDCE target. | ||
| Aluminium | Process setup optimisation | A simulation tool for the recipe (4a) is integrated into the automation system to identify the root cause of production non-conformities. Aligning with the need for automated diagnosis, it is an ART-type application |
| Type implemented: Type 1 | ||
| Analysis: Type 1 does not match the desired configuration for ART target, which would require Type 2 | ||
| 4c is a PDCE when it refers to re-designing recipes, and a DME when it refers to a new design. Collaboration with the human is realised by generating three different recipes, allowing the human to select the most reliable one. In this case, generative AI is applied | ||
| Type implemented: Type 3 (in both cases) | ||
| Analysis: in this case, the implemented type is in line with the DME desirable option | ||
| Case . | Requirement . | Humachine dynamics classification . |
|---|---|---|
| Asphalt | Process setup optimisation | 1a is PDCE when it refers to an existing product for which the current mix recipe is not satisfactory and it is of a DME when about a new product |
| Type implemented: Type 1 (in both cases) | ||
| Analysis: in the current implementation of the mix recipe optimisation several prediction tools are applied but no direct actuation of the automatic decision is ever foreseen, therefore the implemented humachine is always Type 1, while for DME target (new product design) a Type 3 is the desirable option, and for PDCE target (existing product optimisation) the implemented humachine is in line with the desirable option (Type 1) | ||
| 1c is ART since it is deterministically executed every time material is dispatched for defining the target loading temperature | ||
| Type implemented: Type 1 | ||
| Analysis: Type 1 mismatches the desirable option (Type 2) because of the incapability to precisely predict temperature at paving site | ||
| Process r/t control | 1b is DME since technology aims at making diagnoses | |
| Type implemented: Type 1 | ||
| Analysis: Type 1 mismatches the desirable option for a DME target (Type 3) for the automatic reconfiguration. Indeed, ML technologies provide suggestions only, while human validates anomalies and determines mitigation actions | ||
| Steel | Planning | Planning is activated by a human operator (2b) to ensure optimal scrap usage. It is a DME task, since planning tool proposes a solution, and human intervention occurs only if the proposed solution is deemed suboptimal. Moreover, scenario exploration is a typical feature of planning tools, as simulations allowing exploration of alternative scenarios. However, the application of the final plan might be the result of human-MAPE-K cooperation even if systematically human supervised |
| Type implemented: Type 3 | ||
| Analysis: in this case, the implemented type is in line with the DME desirable option | ||
| Process setup optimisation | A simulation tool for the EAF recipe (2a) is integrated into the automation system to identify root causes of production non-conformities. Aligning with the need for automated diagnosis, it is an ART-type application | |
| Type implemented: Type 1 | ||
| Analysis: Type 1 does not match the desired configuration for ART target, which would require Type 2 | ||
| 2c is a PDCE when it refers to re-designing recipes, and a DME when it refers to a new design | ||
| Type implemented: Type 1 (in both cases) | ||
| Analysis: in the current implementation of the design/re-design of a recipe making several prediction tools are applied, but no direct actuation of the automatic decision is foreseen; therefore, the humachine type for the DME target is not the desired one (Type 3), whereas it corresponds to the desirable option and for a PDCE target (Type 1) | ||
| Pharma | Process setup optimisation | The process settling optimisation (3a) can be classified as a PDCE task since the process might be standardised once enough data is collected in all possible working points |
| Type implemented: Type 1 | ||
| Analysis: the implemented type is coherent with the desired one for a PDCE target. | ||
| Process r/t control | An MPC acts continuously on the basis of an incapsulated predictive ML model to regulate a chemical reaction (3b, target classification ART). The controller freezes if the control trim exceeds a tolerance level and, in this case, human intervention is required for correction. In the nominal conditions, the reference generated by the MPC controller is actuated in an unsupervised way. ML model is also applied to automatically define the controller initial settings | |
| Type implemented: Type 2 | ||
| Analysis: the implemented type is coherent with the desired one for an ART target. | ||
| The ML model is also used to diagnose possible OCT misplacements (3c, target classification PDCE). In case of anomaly detection, the OCT positioning remains human actuated | ||
| Type implemented: Type 1 | ||
| Analysis: the implemented type is coherent with the desired one for a PDCE target. | ||
| Aluminium | Process setup optimisation | A simulation tool for the recipe (4a) is integrated into the automation system to identify the root cause of production non-conformities. Aligning with the need for automated diagnosis, it is an ART-type application |
| Type implemented: Type 1 | ||
| Analysis: Type 1 does not match the desired configuration for ART target, which would require Type 2 | ||
| 4c is a PDCE when it refers to re-designing recipes, and a DME when it refers to a new design. Collaboration with the human is realised by generating three different recipes, allowing the human to select the most reliable one. In this case, generative AI is applied | ||
| Type implemented: Type 3 (in both cases) | ||
| Analysis: in this case, the implemented type is in line with the DME desirable option | ||