Page 329 -

12.3 Applications 311

Wafer scanners are complex machines consisting of many building blocks. They
use a photographic process to image nanometric circuit patterns onto a silicon wafer.
Because of competition and fast innovation, the time-to-market is very important
and every new generation of wafer scanners is balancing on the border of what is
technologically possible. As a result, the testing of manufactured wafer scanners is
an important, but also time-consuming, process. Every wafer scanner is tested in
the factory of ASML. When it passes all tests, the wafer scanner is disassembled
and shipped to the customer where the system is re-assembled. At the customer’s
site, the wafer scanner is tested again. Testing is time-consuming and takes several
weeks on both sites. Since time-to-market is very important, ASML is constantly
looking for ways to reduce the time needed to test wafer scanners.
Figure 12.10 shows that the testing of wafer scanners is indeed a Spaghetti pro-
cess [82]. The model was discovered based on an event log containing 154,966
events. The event log contained information about 24 carefully chosen wafer scan-
ners (same type, same circumstances, and having complete logs). The number of
events per case (i.e., the length of the executed test sequence) in this event log ranges
from 2820 to 16,250 events. There are 360 different activities, all identified by four-
letter test codes. Each instance of these 360 activities has a start event and complete
event. Figure 12.10 is based on just the complete events.
ASML also had a so-called reference model describing the way that machines
should be tested. This reference model is at the level of job steps rather than test
codes. However, ASML maintains a mapping from the lower level codes to these
higher level activities. Comparing the reference model and our discovered model
(both at the job step and test code level) revealed interesting differences. Moreover,
using the ProM’s conformance checker we could show that the average fitness was
only fitness(L,N) = 0.375, i.e., less than half of the events can be explained by the
model (Sect. 7.2). When replaying, we discovered many activities that had occurred
but that should not have happened according to the reference model and activities
that should have happened but did not.
Both the discovered process models and the results of conformance checking
showed that process mining can provide new insights that can be used to improve
the management of complex Spaghetti-like processes. We refer to [82]for more
details.

12.3.2.2 Philips Healthcare

Philips Healthcare is one of the leading manufacturers of medical devices, offering
diagnosing imaging systems, healthcare information technology solutions, patient
monitoring systems, and cardiac devices. Like ASML, Philips Healthcare is devel-
oping complex high-tech machines that record massive amounts of events. Since
2007, there has been an ongoing effort to analyze the event logs of these machines
using process mining.
Philips Remote Services (PRS) is a system for the active monitoring of systems
via the Internet. PRS has been established to deliver remote technical support, moni-
toring, diagnostics, application assistance, and other added value services. Low level