Author Contributions
Conceptualization, O.M. and N.K.; methodology, N.K., M.S. (Michalis Skoumperdis), O.M., M.L., I.P., C.K., P.G. and N.K.; software, M.S. (Michalis Skoumperdis), C.K., D.T. (Dimitrios Tzilopoulos), M.O., M.L. and N.T.; validation, M.S. (Michalis Skoumperdis), C.K., D.T. (Dimitrios Tzilopoulos), M.O., M.L., N.T., P.G. and N.K.; formal analysis, M.S. (Michalis Skoumperdis) and N.K.; investigation, M.S. (Michalis Skoumperdis), O.M., A.T., I.P., N.K. and C.K.; resources, O.M., A.T., D.I., D.T. (Dimitrios Tzovaras) and M.S. (Mile Stankovski); data curation, N.T., M.O., D.T. (Dimitrios Tzilopoulos), M.S. (Michalis Skoumperdis) and N.K.; writing—original draft preparation, M.S. (Michalis Skoumperdis); writing—review and editing, M.S. (Michalis Skoumperdis), O.M., D.I. and N.K.; visualization, M.O. and D.T. (Dimitrios Tzilopoulos); supervision, N.K., D.I., D.T. (Dimitrios Tzovaras) and M.S. (Mile Stankovski); project administration, O.M., A.T., D.I., D.T. (Dimitrios Tzovaras) and M.S. (Mile Stankovski); funding acquisition, D.I. and D.T. (Dimitrios Tzovaras). All authors have read and agreed to the published version of the manuscript.
Figure 1.
The cement industry facilities in Greece.
Figure 1.
The cement industry facilities in Greece.
Figure 2.
High-level architecture overview diagram.
Figure 2.
High-level architecture overview diagram.
Figure 3.
Monitoring the results of timed executions in the data reading algorithm.
Figure 3.
Monitoring the results of timed executions in the data reading algorithm.
Figure 4.
The distribution of data by non-manipulated variables in the mill and their value at a particular time point.
Figure 4.
The distribution of data by non-manipulated variables in the mill and their value at a particular time point.
Figure 5.
The distribution of data by non-manipulated variables in the mill and their value at a particular time point.
Figure 5.
The distribution of data by non-manipulated variables in the mill and their value at a particular time point.
Figure 6.
The distribution of data by non-manipulated variables in the kiln and their value at a particular time point.
Figure 6.
The distribution of data by non-manipulated variables in the kiln and their value at a particular time point.
Figure 7.
The flowchart of the platform.
Figure 7.
The flowchart of the platform.
Figure 8.
The flowchart of the AI subsystem.
Figure 8.
The flowchart of the AI subsystem.
Figure 9.
The flowchart of the IDMS subsystem.
Figure 9.
The flowchart of the IDMS subsystem.
Figure 10.
The Distribution of the Kiln Feed parameter.
Figure 10.
The Distribution of the Kiln Feed parameter.
Figure 11.
The Distribution of the Kiln Amps parameter.
Figure 11.
The Distribution of the Kiln Amps parameter.
Figure 12.
The constraints, the process values (PV), the setting points (SP), and the range of constraints (LL, HL) if a manipulated or non-manipulated variable is close to zero (0). The green colour arrow means that the PV is increasing, the red colour arrow that it is decreasing and the yellow colour arrow that it is stable.
Figure 12.
The constraints, the process values (PV), the setting points (SP), and the range of constraints (LL, HL) if a manipulated or non-manipulated variable is close to zero (0). The green colour arrow means that the PV is increasing, the red colour arrow that it is decreasing and the yellow colour arrow that it is stable.
Figure 13.
The option between mill and kiln in the dashboard to determine the corresponding objective function.
Figure 13.
The option between mill and kiln in the dashboard to determine the corresponding objective function.
Figure 14.
Creation of Objective Function via the Dashboard.
Figure 14.
Creation of Objective Function via the Dashboard.
Figure 15.
Creation of Objective Function of the 1st Optimization Strategy via the Dashboard.
Figure 15.
Creation of Objective Function of the 1st Optimization Strategy via the Dashboard.
Figure 16.
Creation of Objective Function of the 2nd Optimization Strategy via the Dashboard.
Figure 16.
Creation of Objective Function of the 2nd Optimization Strategy via the Dashboard.
Figure 17.
Creation of Objective Function of the 3rd Optimization Strategy via the Dashboard.
Figure 17.
Creation of Objective Function of the 3rd Optimization Strategy via the Dashboard.
Figure 18.
Creation of Objective Function of the 4th Optimization Strategy via the Dashboard.
Figure 18.
Creation of Objective Function of the 4th Optimization Strategy via the Dashboard.
Figure 19.
Determination of the DE optimization matrix parameters via the Control Panel.
Figure 19.
Determination of the DE optimization matrix parameters via the Control Panel.
Figure 20.
Recommendations of manipulated variables in the first optimization strategy to reach optimal operation in the cement mill.
Figure 20.
Recommendations of manipulated variables in the first optimization strategy to reach optimal operation in the cement mill.
Figure 21.
Distribution of recommendations of manipulated variables in the first optimization strategy to reach optimal operation in the cement mill.
Figure 21.
Distribution of recommendations of manipulated variables in the first optimization strategy to reach optimal operation in the cement mill.
Figure 22.
Comparison between actual and recommended values of parameters in the first optimization strategy.
Figure 22.
Comparison between actual and recommended values of parameters in the first optimization strategy.
Figure 23.
Comparison between actual and recommended values of parameters in the mill in the first optimization strategy.
Figure 23.
Comparison between actual and recommended values of parameters in the mill in the first optimization strategy.
Figure 24.
Recommendations of manipulated variables in the second optimization strategy to reach optimal operation in the cement mill.
Figure 24.
Recommendations of manipulated variables in the second optimization strategy to reach optimal operation in the cement mill.
Figure 25.
Distribution of recommendations of manipulated variables in the second optimization strategy to reach optimal operation in the cement mill.
Figure 25.
Distribution of recommendations of manipulated variables in the second optimization strategy to reach optimal operation in the cement mill.
Figure 26.
Comparison between actual and recommended values of parameters in the second optimization strategy.
Figure 26.
Comparison between actual and recommended values of parameters in the second optimization strategy.
Figure 27.
Comparison between actual and recommended values of parameters in the mill in the second optimization strategy.
Figure 27.
Comparison between actual and recommended values of parameters in the mill in the second optimization strategy.
Figure 28.
Recommendations of manipulated variables in third optimization strategy to reach optimal operation in the cement mill.
Figure 28.
Recommendations of manipulated variables in third optimization strategy to reach optimal operation in the cement mill.
Figure 29.
Distribution of recommendations of manipulated variables in third optimization strategy to reach optimal operation in the cement mill.
Figure 29.
Distribution of recommendations of manipulated variables in third optimization strategy to reach optimal operation in the cement mill.
Figure 30.
Comparison between actual and recommended values of parameters in third optimization strategy.
Figure 30.
Comparison between actual and recommended values of parameters in third optimization strategy.
Figure 31.
Comparison between actual and recommended values of parameters in the mill in third optimization strategy.
Figure 31.
Comparison between actual and recommended values of parameters in the mill in third optimization strategy.
Figure 32.
Recommendations of manipulated variables in fourth optimization strategy to reach optimal operation in the cement mill.
Figure 32.
Recommendations of manipulated variables in fourth optimization strategy to reach optimal operation in the cement mill.
Figure 33.
Recommendations of manipulated variables in fourth optimization strategy to reach optimal operation in the cement mill.
Figure 33.
Recommendations of manipulated variables in fourth optimization strategy to reach optimal operation in the cement mill.
Figure 34.
Comparison between actual and recommended values of parameters in fourth optimization strategy.
Figure 34.
Comparison between actual and recommended values of parameters in fourth optimization strategy.
Figure 35.
Comparison between actual and recommended values of parameters in the mill in fourth optimization strategy.
Figure 35.
Comparison between actual and recommended values of parameters in the mill in fourth optimization strategy.
Figure 36.
The process of creating an AI model by the user.
Figure 36.
The process of creating an AI model by the user.
Figure 37.
Comparison of the active value of the objective function and the function determined by the DE when the active AI model was created by the user.
Figure 37.
Comparison of the active value of the objective function and the function determined by the DE when the active AI model was created by the user.
Figure 38.
Comparison between actual and recommended values of parameters in the mill in fourth optimization strategy.
Figure 38.
Comparison between actual and recommended values of parameters in the mill in fourth optimization strategy.
Figure 39.
Creation of Objective Function of the 1st Optimization Strategy via the Dashboard.
Figure 39.
Creation of Objective Function of the 1st Optimization Strategy via the Dashboard.
Figure 40.
Creation of Objective Function of the 2nd Optimization Strategy via the Dashboard.
Figure 40.
Creation of Objective Function of the 2nd Optimization Strategy via the Dashboard.
Figure 41.
Creation of Objective Function of the 3rd Optimization Strategy via the Dashboard.
Figure 41.
Creation of Objective Function of the 3rd Optimization Strategy via the Dashboard.
Figure 42.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the first optimization strategy.
Figure 42.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the first optimization strategy.
Figure 43.
The quantitative difference between the active value of the objective function and the value explained by the DE in the first optimization strategy.
Figure 43.
The quantitative difference between the active value of the objective function and the value explained by the DE in the first optimization strategy.
Figure 44.
The distribution of system’s recommendations in kiln in the first optimization strategy.
Figure 44.
The distribution of system’s recommendations in kiln in the first optimization strategy.
Figure 45.
Comparison between actual and recommended values of parameters in the kiln in the first optimization strategy.
Figure 45.
Comparison between actual and recommended values of parameters in the kiln in the first optimization strategy.
Figure 46.
The quantitative difference between the active value of the objective function and the value explained by the DE in the second optimization strategy.
Figure 46.
The quantitative difference between the active value of the objective function and the value explained by the DE in the second optimization strategy.
Figure 47.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the second optimization strategy.
Figure 47.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the second optimization strategy.
Figure 48.
The distribution of system recommendations in the kiln in the second optimization strategy.
Figure 48.
The distribution of system recommendations in the kiln in the second optimization strategy.
Figure 49.
Comparison between actual and recommended values of parameters in the kiln in the second optimization strategy.
Figure 49.
Comparison between actual and recommended values of parameters in the kiln in the second optimization strategy.
Figure 50.
The quantitative difference between the active value of the objective function and the value explained by the DE in the third optimization strategy.
Figure 50.
The quantitative difference between the active value of the objective function and the value explained by the DE in the third optimization strategy.
Figure 51.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the third optimization strategy.
Figure 51.
Proposed optimal values for the manipulated variables to attain optimal operation in the cement kiln in the third optimization strategy.
Figure 52.
The distribution of system recommendations in the kiln in the third optimization strategy.
Figure 52.
The distribution of system recommendations in the kiln in the third optimization strategy.
Figure 53.
Comparison between actual and recommended values of parameters in the kiln in the third optimization strategy.
Figure 53.
Comparison between actual and recommended values of parameters in the kiln in the third optimization strategy.
Table 1.
Overview of the relevant works. The checkmark point indicates that the method is relevant to the category indicated in the corresponding column.
Table 1.
Overview of the relevant works. The checkmark point indicates that the method is relevant to the category indicated in the corresponding column.
Reference | Cement Mill Optimization | Cement Kiln Optimization | Optimization of Other Functionality | Mathematical Optimization Method | Improvement of Architecture and/or Control Process | Method |
---|
[7] | - | - | ✓ | ✓ | - | GA |
[18] | - | ✓ | - | ✓ | - | ANFIS-GA |
[20] | - | ✓ | ✓ | - | ✓ | HRSO |
[21] | - | ✓ | ✓ | - | ✓ | PM-NOx concentration |
[22] | - | ✓ | ✓ | - | ✓ | CFD |
[23] | ✓ | ✓ | ✓ | ✓ | - | GA |
[24] | - | ✓ | ✓ | - | ✓ | RMCM |
[25] | - | - | ✓ | ✓ | ✓ | Introducing nano-Si |
[26] | - | - | ✓ | ✓ | ✓ | Creation of DA-PEG |
[27] | - | ✓ | - | ✓ | - | PC - SCN - SSA |
[28] | ✓ | - | ✓ | ✓ | - | MOBAS |
[29] | ✓ | - | ✓ | ✓ | - | BBD |
[30] | ✓ | ✓ | ✓ | ✓ | - | BPNN - MOBAS |
[31] | ✓ | - | ✓ | ✓ | - | PSOA |
[32] | ✓ | - | ✓ | ✓ | - | BO-LightGBM |
[33] | ✓ | ✓ | ✓ | ✓ | - | BDA |
[34] | ✓ | ✓ | ✓ | ✓ | - | BDA |
[35] | ✓ | ✓ | ✓ | - | ✓ | PLC |
[36] | ✓ | - | - | - | ✓ | APC |
[37] | ✓ | ✓ | ✓ | - | ✓ | IS |
[38] | - | ✓ | - | ✓ | - | SAAS, AI, APC |
[39] | ✓ | - | - | ✓ | - | AI, DE |
Our Approach | ✓ | ✓ | - | ✓ | - | AI, DE |
Table 2.
The selected features in cement mill for each variable.
Table 2.
The selected features in cement mill for each variable.
Dependent Variable Y | Independent Variables | Manipulated Variables |
---|
Mill Motor | Grinding Pressure, Separator Speed, Mill Feed, Water Flow, Mill Inlet Pressure, Limestone%, Pozzolana%, FlyAsh%, Grinding Aid PV, Grinding Layer Roller | Grinding Pressure, Separator Speed, Mill Feed, Water Flow, Mill Inlet Pressure |
Mill Differential Pressure | Mill Feed, Mill Outlet Pressure, Bag Filter, Mill Inlet Pressure, Separator Speed, Gypsum% | Mill Feed, Mill Outlet Pressure, Mill Inlet Pressure, Separator Speed |
Separator Motor | Mill Feed, Separator Speed, Mill Outlet Pressure, Grinding Pressure, Mill Inlet Temperature, Limestone%, Pozzolana%, Fly Ash%, Grinding Layer Roller | |
Mill Exit Temperature | Mill Feed, Grinding Pressure, Separator Speed, Mill Inlet Temperature, Mill Inlet Pressure, Limestone%, Pozzolana%, FlyAsh% | Mill Feed, Grinding Pressure, Separator Speed, Mill Inlet Temperature, Mill Inlet Pressure |
Environmental Dust | Separator Speed, Bag Filter, Limestone% | Separator Speed |
Blaine | Mill Outlet Pressure, Limestone%, FlyAsh% | Mill Outlet Pressure |
Residue | Mill Inlet Pressure, Mill Outlet Pressure, Water Flow, Limestone% | Mill Inlet Pressure, Mill Outlet Pressure, Water Flow |
Mill Vibrations | Mill Feed, Mill Inlet Pressure, Water Flow, Separator Speed, Grinding Pressure, Limestone% | Mill Feed, Mill Inlet Pressure, Water Flow, Separator Speed, Grinding Pressure |
Table 3.
The selected features in Kiln for each variable.
Table 3.
The selected features in Kiln for each variable.
Dependent Variable Y | Independent Variables | Manipulated Variables |
---|
Calculated NOx | Heat Main Burner, Clinker O, Kiln Feed LSF, Total Feed, Preheater, Solid Fuel Feed | Preheater, Solid Fuel Feed, Total Feed |
Kiln Amps | Total Air Flow, Secondary Air Temp, Kiln Vortex Temp, Solid Fuel Feed, Kiln Inlet Press, Total Feed, Preheater | Preheater, Solid Fuel Feed, Total Feed |
Preheater CO | Kiln Inlet Press, Press Transport Air, Press MAS Air, Total Air Flow, Clinker O, Secondary Air Temp, Total Feed, Preheater, Solid Fuel Feed | Preheater, Solid Fuel Feed, Total Feed |
Table 4.
The results of supervised learning models for the cement mill.
Table 4.
The results of supervised learning models for the cement mill.
Variable | Model Type | Number of Estimators | Max Depth | Number of Leaves | NRMSE |
---|
Mill Motor | RF | 10 | 5 | - | 0.34 |
Mill Differential Pressure | LGBM | - | 10 | 10 | 0.23 |
Separator Motor | LGBM | - | 5 | 12 | 0.49 |
Mill Exit Temperature | XGBoost | 50 | 3 | - | 0.35 |
Environmental Dust | GBR | 50 | 3 | - | 0.89 |
Blaine | GBR | 50 | 3 | - | 0.67 |
Residue | TTR | - | - | - | 0.78 |
Mill Vibrations | RF | 50 | 5 | - | 0.69 |
Table 5.
The results of supervised learning models for the kiln.
Table 5.
The results of supervised learning models for the kiln.
Variable | Model Type | Fit Intercept | Number of Estimators | Max Depth | Hidden Layer Sizes | Number of Leaves | NRMSE |
---|
Calculated NOx | MLP | - | - | - | (5,) | - | 0.95 |
Kiln Amps | XGBoost | - | 50 | 3 | - | - | 0.14 |
Preheater CO | LGBM | - | - | 5 | - | 8 | 0.96 |
Table 6.
Results of the cement mill’s supervised learning models when the data are clustered.
Table 6.
Results of the cement mill’s supervised learning models when the data are clustered.
Variable | Model Type | Number of Estimators | Max Depth | Number of Leaves | Hidden Layer Sizes | NRMSE |
---|
Mill Motor | RF | 50 | 10 | - | - | 0.40 |
Mill Differential Pressure | MLP | - | - | - | (5,) | 0.21 |
Separator Motor | XGBoost | 50 | 3 | - | - | 0.53 |
Mill Exit Temperature | XGBoost | 50 | 3 | - | - | 0.37 |
Environmental Dust | LGBM | - | 5 | 8 | - | 0.84 |
Blaine | MLP | - | - | - | (5,) | 0.68 |
Residue | TTR | - | - | - | - | 0.76 |
Mill Vibrations | LGBM | - | 5 | 10 | - | 0.70 |
Table 7.
Results of the cement kiln’s supervised learning models when the data are clustered.
Table 7.
Results of the cement kiln’s supervised learning models when the data are clustered.
Variable | Model Type | Number of Estimators | Max Depth | Number of Leaves | NRMSE |
---|
Calculated NOx | TTR | - | - | - | 0.99 |
Kiln Amps | XGBoost | 50 | 3 | - | 0.21 |
Preheater CO | GBR | 50 | 3 | - | 0.98 |
Table 8.
Optimization strategies and weights of each factor in the cement mill.
Table 8.
Optimization strategies and weights of each factor in the cement mill.
Optimization Strategy | Main Target | Mill Feed Weight | Mill KW Weight | Separator Power Weight | Vibrations Weight | Mill DP Weight | Environ. Dust Weight | Mill Exit Temperature Weight | Blaine Weight | Residue Weight |
---|
1st | Combined Optimal | 1 (increase) | (decrease) | (decrease) | (decrease) | (increase) | (decrease) | (increase) | 10 (Target) | 10 (Target) |
2nd | Energy Optimal | (increase) | (decrease) | (decrease) | - | - | - | - | - | - |
3rd | Environmental Impact | - | (decrease) | (decrease) | - | - | (decrease) | (increase) | - | - |
4th | Production Cost | (increase) | (decrease) | (decrease) | - | - | - | - | (Target) | (Target) |
Table 9.
Optimization Strategies and weight of each factor in cement kiln.
Table 9.
Optimization Strategies and weight of each factor in cement kiln.
Optimization Strategy | Main Target | Calculated NOx | Kiln Amps | Preheater CO | Total Feed | Preheater | Solid Fuel Feed |
---|
1st | Combined Optimal | (decrease) | (decrease) | (decrease) | 10 (increase) | (decrease) | 10 (decrease) |
2nd | Production Cost | - | (decrease) | - | 10 (increase) | - | (decrease) |
3rd | Environmental Impact | (decrease) | (decrease) | (decrease) | - | - | (decrease) |