460
O. Ogorodnyk et al.
During the experiment “ENGEL insert 130” vertical injection molding machine
with CC300 control unit has been used and 47 machine and process parameters from
the built-in by the machine manufacturer sensors were logged. The logging system
has been developed using the Python programming language to be able to establish a
connection with the machine and access all the necessary process parameter values.
Step 2: Data pre-processing. The logged experimental data included values of
different machine and process parameters for every 0.5 s during every production
cycle. However, some of the signals vary less than the others during one production
cycle but change a lot from one cycle to another. As a result, the logged data needed
certain “pre-processing” before any further use. Pre-processing resulted in one value
of a process parameter per production cycle, some of the parameter values were
averaged, some were taken the maximum or minimum value of, etc. An example of
the structure of the data after the “pre-processing” is shown in Table 32.1. The top
row includes names of the logged parameters and each following row corresponds
to one data sample. The “pre-processing” has been done using scripts developed in
the Python programming language.
Step 3: Quality data collection. After the production of the focus parts, the quality
data needed to be obtained. Since dimensional consistency is one of the important
criteria for the part’s acceptance, the produced dogbone specimens were measured
using ZEISS DuraMax coordinate measuring machine (DuraMax 2019).
The measurements have been done according to the “ISO 16012: Plastics—Determination of linear dimensions of the test specimens” (ISO 2015), the accuracy of the
machine in the temperature 18–22 °C is ±2.4 µm. The precision error doesn’t exceed
±0.02 mm for dimensions <10 (thickness) and ±0.1 (width) for ≥ 10. Figure 32.1
depicts one of the specimens being measured by the coordinate measuring machine.
Step 4: Important parameters selection. After obtaining both process and quality
data, it was necessary to understand which parameters should be included in the
quality prediction model. First, all parameters that had constant values during all
160 runs were excluded.
Secondly, six parameters (machine time, shot counter, good parts counter, bad
parts counter, parts counter and machine date) were eliminated, because they do not
contain any necessary information about the process.
Thirdly, feature selection algorithms were applied to define which parameters
out of 25 that were left are the most influential. According to the Correlation-based
feature selection algorithm with the Best First search algorithm only seven parameters
should be included in the model: cushion after holding pressure, plasticizing time,
holding pressure time, cushion smallest value, injection work, holding pressure and
Table 32.1 Experimental data
ID
Max screw speed
Pressure at switchover
Cushion size after holding pressure
1
239.18
1058.69
4.90
2
239.40
1059.35
4.92
3
239.52
1059.62
4.98
O. Ogorodnyk et al.
During the experiment “ENGEL insert 130” vertical injection molding machine
with CC300 control unit has been used and 47 machine and process parameters from
the built-in by the machine manufacturer sensors were logged. The logging system
has been developed using the Python programming language to be able to establish a
connection with the machine and access all the necessary process parameter values.
Step 2: Data pre-processing. The logged experimental data included values of
different machine and process parameters for every 0.5 s during every production
cycle. However, some of the signals vary less than the others during one production
cycle but change a lot from one cycle to another. As a result, the logged data needed
certain “pre-processing” before any further use. Pre-processing resulted in one value
of a process parameter per production cycle, some of the parameter values were
averaged, some were taken the maximum or minimum value of, etc. An example of
the structure of the data after the “pre-processing” is shown in Table 32.1. The top
row includes names of the logged parameters and each following row corresponds
to one data sample. The “pre-processing” has been done using scripts developed in
the Python programming language.
Step 3: Quality data collection. After the production of the focus parts, the quality
data needed to be obtained. Since dimensional consistency is one of the important
criteria for the part’s acceptance, the produced dogbone specimens were measured
using ZEISS DuraMax coordinate measuring machine (DuraMax 2019).
The measurements have been done according to the “ISO 16012: Plastics—Determination of linear dimensions of the test specimens” (ISO 2015), the accuracy of the
machine in the temperature 18–22 °C is ±2.4 µm. The precision error doesn’t exceed
±0.02 mm for dimensions <10 (thickness) and ±0.1 (width) for ≥ 10. Figure 32.1
depicts one of the specimens being measured by the coordinate measuring machine.
Step 4: Important parameters selection. After obtaining both process and quality
data, it was necessary to understand which parameters should be included in the
quality prediction model. First, all parameters that had constant values during all
160 runs were excluded.
Secondly, six parameters (machine time, shot counter, good parts counter, bad
parts counter, parts counter and machine date) were eliminated, because they do not
contain any necessary information about the process.
Thirdly, feature selection algorithms were applied to define which parameters
out of 25 that were left are the most influential. According to the Correlation-based
feature selection algorithm with the Best First search algorithm only seven parameters
should be included in the model: cushion after holding pressure, plasticizing time,
holding pressure time, cushion smallest value, injection work, holding pressure and
Table 32.1 Experimental data
ID
Max screw speed
Pressure at switchover
Cushion size after holding pressure
1
239.18
1058.69
4.90
2
239.40
1059.35
4.92
3
239.52
1059.62
4.98