(140b) “Green” Solvents: Multi-Criteria Knowledge-Based and Statistical Approaches for Their Evaluation

?Green? solvents: Multi-criteria
knowledge-based and statistical approaches for their evaluation

Stavros Papadokonstantakis, Tianhe Liu, Alireza
Banimostafa, Konrad Hungerbühler

Institute of Chemical- and Bioengineering, Swiss
Federal Institute of Technology (ETH) Zurich, Switzerland

Sustainable process design has
been recognized as one of the key research challenges for process systems
engineering. In fact, the experience gained so far has pointed out that the
applicability of sustainability principles can be more advantageous in earlier
phases of process design characterized by more degrees of freedom for decision
making. Typical decisions made in these early process design phases involve the
selection of chemical synthesis path, chemical auxiliaries, unit operation
conditions up to a basic flowsheet design.

Solvents belong to chemical
auxiliaries of particular interest for the chemical industry due to their
multi-functional use (e.g., to achieve desired reaction rates while protecting
from the reaction exothermicity, to ensure solid product quality requirements
through effective crystallization, to perform liquid-liquid separations or azeotropic
distillations etc.) and their large amounts (e.g., in fine-chemical and
pharmaceutical production large amounts are used per mass of final products). Historically,
solvents have been designed to maximize technical performance and minimize
cost. But nowadays, an effort to find ?green' solvents gains interest, for
replacing conventional solvents with environmentally benign substitutes while,
of course, maintaining their effectiveness for the intended use. In this
context, the concept of ??green'' solvents expresses the goal to minimize the
environmental impact resulting from their use in chemical production, both from
a life cycle impact assessment (LCIA) and a safety, health and environmental
(SHE) hazard assessment point of view.

Several shortcut, index-based
methodologies have been proposed for evaluating the relevant LCIA and SHE
effects, both for independent substances and for chemical production processes.
These index-based frameworks encapsulate prior knowledge regarding main SHE
hazards, cradle-to-gate resources consumption, and emission related effects.
The indices are typically defined on the basis of substance properties, derived
from available databases or estimated using available techniques (thermodynamic
property estimation methods, QSAR models, etc.), and process features derived
either from a preliminary process flowsheet or even on the basis of the chemical
synthesis path. On the one hand, the simplicity of the index-based approaches
makes them attractive, especially for early stages of process design lacking
detailed process information, but on the other hand they face criticism
regarding limited coverage of considered effects, subjectivities for the
calculation of aggregated end-point indices, and unknown resolution of the
scaled final scores according to which the ranking of process alternatives is
inferred.

Especially for the problem of
subjective aggregation to end-point scores principal component analysis (PCA)
has been recently proposed as a potential remedy. In this approach the rows of
the PCA input matrix refer to the evaluation objects (e.g., solvents, chemical
process steps, chemical process sections or overall chemical processes) and the
columns refer to metrics for their assessment (e.g., LCIA and/or SHE hazard
assessment indicators). The evaluation objects are ranked according to the
obtained principal component (PC) scores, which in the simplest PCA version can
be viewed as linear combinations of the assessment metrics. The weights of
these linear combinations express the importance of the original assessment
metric for a specific PC and are derived in such a way that the first PC covers
most of the variability of the PCA input matrix, the second PC most of the
remaining variability under the condition that it is orthogonal to the first
one, and so on. The concept behind this PCA-based method, e.g., when it is used
for the assessment of substances, is that dependencies must exist between the
assessment metrics, since the molecular structure of a substance should, in
theory, provide (although most of the times via unknown functional
relationships) all necessary information for the assessment metrics. The
PCA-based method will, therefore, detect these dependencies in a theoretically
sound way and this information will be used instead of a priori
weighting schemes for the diverse metrics. In this way, two objectives are
achieved simultaneously, i.e., the inter-relationships between the diverse metrics
reveal the ?true? dimensionality of the multi-criteria assessment and the level
of multi-criteria similarity among individual solvents can be derived. Moreover,
the same PCA-based concept can be used in an expectation-maximization iterative
procedure for filling data gaps, typically occurring in the assessment metrics.

The proposed framework is
demonstrated (see Figure 1) on the basis of 138 organic solvents used in
industrial applications. The performance of the PCA-based method for filling
data gaps is compared with other approaches based on priority settings between
substance properties and reveals in which extent this should be preferred
(i.e., relative amount and patterns of data gaps). Moreover, two case studies
are presented in which the results of the PCA-based framework are compared to
the results of other studies (from the Safety and Environmental Technology
group in ETHZ and GlaxoSmithKline, respectively) using different index-based methods
with a priori postulated weighting schemes. The results indicate that
the presented framework is useful for covering most of the information contained
in the other index-based methods for the evaluation of ?green? solvents, while
being inherently flexible with respect to the number and type of assessment
metrics, as well as to the existence of data gaps. Finally, the present study
discusses the advantages and limitations of the PCA-based approach in terms of
simplicity, interpretability and statistical robustness.

Figure 1:
Integrating a PCA-based approach for the evaluation of ?green solvents? and
comparison with existing index-based approaches.