*3.3. Process Fingerprint Analysis*

In the same way as for the "product fingerprint", a set of "process fingerprint" candidates were extracted from the machine process monitoring and regulation signals. The goal was to verify which can act as indicators of the overall product quality, especially for the functional μ-pillar features. The time-resolved machine data were used to extract two type of indicators:


#### 3.3.1. Process Fingerprint Based on Indicators of Type 1

As already discussed in Section 2.5.1, the machine controller records the injection speed and the pressure time series and transient data during the process for every consecutive cycle. The deviation of those signals from the initial reference (1st signal per run) is used for the calculation of the deviation-based "process fingerprints" (Type 1). When this type of indicator is considered, the "fingerprints" as well as the dataset's values are standardized in order to be compared. Figure 16 provides an example of the trends that exist between the standardized mold-part deviation measurement and the standardized "process fingerprints" candidate values. It can be seen, particularly for experimental run 16, that not all deviation datasets follow the same trend of the "process fingerprints" candidates. However, the same fingerprints and dataset trends exist for both the nominal (see Figure 16 top) and cross correlated aligned signals (see Figure 16 bottom). It can be seen that "Workdev-InjPr" and "Erwork-InjPr" follow the exact trend with the dataset "C2PP2". Analogously "ISE-InjPr" follows a similar trend. Moreover, the dataset of position RP2 in Cavity 2 (C2RP2) follows a similar trend to fingerprint candidate "ShiftXcorr-InjPr". A similar trend can be observed between the "process fingerprint" candidate "WarpDis-InjPr" and the dataset of position C1PP3.

**Figure 16.** Example of process fingerprint candidates to measurand trends for experimental run 16 based on (**a**) nominal signals and (**b**) cross correlated signals. The legend of the graphs denotes both the measurand datasets (i.e., C1PP1: Cavity 1–Position PP1) and the deviation based (Type 1) "process fingerprints".

When the whole experimental space is considered, the same dataset trends were not always in agreement with the trends of the same candidates. Figure 17 illustrates the occurrence of similar "process fingerprint" trends to the measurement datasets of each experimental run. For example, significant trends between the measurement datasets and the candidate "WarpDis-InjPr" occur a maximum of six times (i.e., six datasets) for Runs 7 and 15 where the Tmelt parameter is kept on the low level. The second process fingerprint candidate occurrence is "Erwork-InjSp" with five times for Run 9 and four times for Run 1. Instead, "process fingerprint" candidates such as "Workdev-InjPr", "ISE-InjPr", and "ISE-InjSp" have less similar trends to the measurement datasets from the same run, even though they appear to have similar trends to measurement datasets from most of the experimental runs.


**Figure 17.** Process fingerprint candidate trend occurrence per Run.

As a conclusion, "process fingerprints" "Workdev-InjPr", "ISE-InjPr", and "ISE-InjSp" together with "WarpDis-InjPr" are considered suitable for the quality control of the pillar μ-features in most of the examined experimental space. However, their correlation and trend are directly dependent on each of the treatments' process parameter combination.

#### 3.3.2. Process Fingerprint Based on Indicators of Type 2

The second type of "process fingerprint" candidates originates from each signal individually. To examine the suitability of signal integrals and signal power to serve as "process fingerprint" candidates, a correlation analysis to respective measurement datasets was conducted with the correlation coefficients |ρ| to be presented in Figure 18a for Cavity 1 and Figure 18b for Cavity 2. The maximum correlation coefficient (|ρ| = 0.436, indicating a moderate correlation) values occur for the combination I.InjSp/C2PP2 (integral of injection speed signal vs. the dataset in position C2PP2, in the middle of the part). The rest of the combinations had weak correlation: they exhibited |ρ| values lower than 0.4. For this reason, the integral and power of the injection pressure and speed signals originating from the IM machine were not considered suitable "process fingerprint" candidates for the quality control and assurance for μ-pillar structured molded components for the particular application.

**Figure 18.** Pearson correlation coefficient plots of measurands to the pillar "product fingerprint" positioned on the runner of the molding (**a**) in Cavity 1 and (**b**) in Cavity 2.

#### **4. Conclusions**

A new approach towards process monitoring and fast-integrated quality assurance of injection molded microstructured components based on product and process fingerprints was presented and validated in this paper. The concept is examined on two parallel tracks. Micro pillars were positioned on the runner before each cavity to serve as "product fingerprints" and the process controlling signals were collected to extract "process fingerprint" candidates. The suitable fingerprints were selected after a sensitivity and correlation analysis was conducted to assess their sensitivity to process variation and correlation, respectively. As far as the product quality assurance was concerned, the replication quality of the μ-pillars was assessed using 3D scanning focus variation microscopy (i.e., off-line metrology). For the process monitoring, the signals generated by the machine regulation embedded sensors were used to extract the time-resolved data. Summarizing the key findings of the research, the following conclusions can be drawn:


Future work will aim at the validation of the presented concept, enriched with data acquired from in-mold temperature and pressure sensors. Furthermore, the assessment of product and process fingerprints performance robustness will be carried out in longer manufacturing runs emulating an actual production environment.

**Author Contributions:** N.G., G.T., and Y.Z. conceived and designed the experiments; N.G. performed the experiments and measurements; N.G. analyzed the data; G.T. and Y.Z. consulted on the work; N.G. wrote the paper; G.T. and Y.Z. revised the paper.

**Funding:** This research was funded by Innovation Fund Denmark grant number 3067-00001B and the APC was funded by the Technical University of Denmark.

**Acknowledgments:** This paper reports work undertaken within the framework of the project MADE (Manufacturing Academy of Denmark, http://en.made.dk/) Work Package 3 "3D Print and New Production Processes" of the Research Platform MADE SPIR (Strategic Platform for Innovation and Research, http://en.made.dk/spir/). MADE is a collaborative research project supported both by the Danish Manufacturing Industry and by the Innovation Fund Denmark (https://innovationsfonden.dk/en). MADE and Innovation Fund Denmark are thanked for providing financial support to the PhD project "Precision Injection Moulding of Micro Features using Integrated Process/Product Quality Assurance".

**Conflicts of Interest:** The authors declare no conflict of interest.
