Knowledge-based Analysis of Residential Air Quality

Aaron Hunter and Rodrigo Mora

British Columbia Institute of Technology, 3700 Willingdon Avenue, Burnaby, Canada

Keywords: AI Applications, Knowledge Representation, Building Science.

Abstract: This paper proposes an approach for residential air quality investigations (IAQ), building on a knowledge-

based theory of building science for systems integration. We present a case study related to the diagnosis of

an air quality problem in a residential building, and we suggest that a logic-based formalization can help direct

investigators towards solutions. This is a problem of significant practical importance, which has not been

specifically addressed in the AI research community. It is envisioned that a formal methodology could

improve storage and retrieval of archival information, and it could be used as a reasoning engine for diagnosis.

1 INTRODUCTION

Building Science is the integrated study of building

performance. This is an evolving discipline

concerned with a wide range of issues, including

indoor air quality (IAQ), heating systems, and

construction materials (Mora et al. 2011). In this

paper, we suggest that a knowledge-based

formalization of building performance could lead to

the development of automated reasoning tools to

support practical investigations.

We focus on IAQ investigations. We suggest that

there are at least two ways in which formal methods

can inform IAQ investigations. First, a formal

ontology representing the domain can clarify exactly

how different components of the system interact.

Second, given such an ontology, we can

automatically diagnose problems through formal

reasoning. However, the reasoning required is non-

monotonic because conclusions need to be retracted

as new information is obtained. This means that a full

treatment of the problem may require a precise model

of ontology evolution. We propose a solution based

on formal models of belief change.

This is a preliminary position paper, outlining the

advantages and challenges related to formal

reasoning for IAQ investigations. The goal is to

outline possible solutions, to be explored in future

work in collaboration with building scientists.

2 MOTIVATION

2.1 Building Science Integrated

Systems

Air quality impacts human health as well as climate

change, due to issues of power consumption.

However, the factors influencing air quality can be

complex and difficult to measure. Building Science

Integrated Systems (BSYS) is the knowledge-based

study of building systems, with the goal of developing

practical systems to assist in reasoning about

problems with building performance.

BSYS research is case driven, using case studies

to discover the knowledge used by professionals to

diagnose and solve building problems. IAQ

investigations are carried out by experts that use

working hypotheses to limit the solution space, and

also use case-based reasoning associating systems to

potential causes (de Mast 2011). IAQ problems are

usually identified by occupants’ complaints related to

odors or breathing problems. An investigation then

consists of screening for possible sources of the

problem, and then testing air quality.

Figure 1 illustrates the high-level structure of

reasoning involved in IAQ investigations. We remark

that, from a formal perspective, the process involves

abductive reasoning, deductive reasoning and

diagnosis.

Hunter, A. and Mora, R.

Knowledge-based Analysis of Residential Air Quality.

DOI: 10.5220/0009102908010805

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 2, pages 801-805

ISBN: 978-989-758-395-7; ISSN: 2184-433X

801

Figure 1: Reasoning about Air Quality.

2.1 Case Study

In this section, we present a case study that will be

used to motivate and guide the development of our

formal approach. The case study is a microbial

investigation of a house with a child with asthma like

symptoms. This case study is an informal description

of an actual investigation carried out in British

Columbia, Canada. The investigation is concerned

with asthma symptoms brought on by indoor air

quality, as discussed in (Morgan 2004).

The climate around the house is characterized by

mild winters with long periods of rain with little sun,

and mild summers. The envelope cladding is cedar

which is strapped to a wood-frame structural wall.

The house has a ventilated attic and an unvented

crawl space, which are outside of the conditioned

envelope. Electric baseboards are used for heating

and ventilation is uncontrolled, through the envelope

cracks. The house is occupied by a mother and her

child. We present the results of the two preliminary

steps of an IAQ investigation for this case study.

Screening:

The goal of screening is to gather quick evidence to

test the hypothesis developed as part of the

inspection. In this case the initial hypothesis is that

the house has “normal and typical” types and amounts

of airborne mold spores. This hypothesis could be

disproved by observing visible mold.

Sampling/Testing and Monitoring:

From the screening, no mold colonies are observed in

the house. Air samples can be taken to verify that the

dynamic indoor conditions are within healthy limits,

not excessively damp or dry. Laboratory testing can be

done to check for spore counts in various parts of the

house, such as the attic or basement. In this particular

case, suppose that we find a higher concentration of

sports in an unventilated crawl space. Based on this, it

can be confirmed that there is a high risk of mold

spores, that will migrate into the house.

Integrated Solution:

In this case study, the following solution may be

proposed:

1. Contaminant Removal - clean out existing

mold from the attic and crawlspace.

2. Dispersion Control – fix air tightness of attic

and crawlspace.

3. Clean Fresh Air Provision – provide

controlled ventilation.

This simple case study serves to illustrate the

process that an investigator may follow in an IAQ

investigation. Note that the investigation requires

screening and testing various possible explanations

for a symptom; these tests may be expensive. Note

also that the proposed solution is occupant centered.

As such, it considers all the house systems that can

possibly play a role in affecting the occupant

exposure to microbial contaminants in the air. The

solution involves three measures to mitigate the

exposure of any microbial source by a receptor inside

the house.

The fundamental point for our purposes is that the

investigator follows a relatively predictable reasoning

process. Given the symptoms, we look to verify find

possible causes. However, given the cost of the

relevant tests and the importance of the solution to the

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

802

occupants, any AI intervention to guide the process

could be potentially valuable.

3 PRELIMINARIES

3.1 Ontologies

An ontology is a formal specification of an

application domain, that makes explicit the

individuals in the domain as well as the relationships

between the individuals. Ontology languages provide

a uniform, structured vocabulary for representing and

reasoning about objects in a wide range of

applications. One of the most popular ontology

languages is the Web Ontology Language (OWL)

(Motik et. al 2012), which was originally developed

for the Semantic Web but has found application in a

wide range of applications.

OWL is built on top of the Resource Description

Framework (RDF). The syntax of RDF is based on

the idea of an RDF triple, which consists of a property

that is applied to two individual objects. Hence, RDF

is able to express statements of the form

MotherOf (Mary, John)

to indicate that the individual Mary is the mother of

the individual John.

OWL extends RDF with additional vocabulary for

describing relationships between concepts, properties

and individuals. The portion of the ontology that

defines roles and concepts is called the T-Box; the

portion of the ontology that defines individuals is

called the A-Box. It is significant to note that ontology

languages (such as OWL) are first-order logics.

ropositional logic is not expressive enough to

explicitly capture the distinction between statements

about concepts and statements about individuals.

3.2 Answer Set Programming

Answer Set Programming (ASP) is a logical

framework for Knowledge Representation based on

the notion of finding non-circular solution to sets of

constraints (Baral 2003). The constraints are written

in the form



where A and B are propositional variables. This rule

is read as an implication, that B is true whenever A is

true. An answer set for a set of rules is a minimal set

of propositional variables that satisfies the rules in a

non-circular manner.

Answer set programming has proven to be one of

the most effective declarative models for non-

monotonic reasoning, and a number of powerful

solvers for answer set programming have been

developed.

3.3 Belief Revision

Belief revision refers to the process in which an agent

must incorporate some new information together with

some pre-existing beliefs. One of the most influential

approaches to belief revision is the AGM approach

(Alchourron 1985). In the AGM approach, the state

of the world is represented by an interpretation of a

propositional signature. The beliefs of an agent are

represented by a set of interpretations K, intuitive the

set of states that an agent considers possible. An

AGM revision operator is defined syntactically

through a set of postulates, and it has been shown that

every AGM revision operator ∗ works as follows.1

Given a belief state K and a formula φ for revision, ∗

maps K to a total-preorder ≺ over states in which the

minimal elements are precisely the elements of K. We

think of ≺ as a “plausibility ordering” that indicates

the most plausible alternatives to the agent’s initial

beliefs. The revised belief state K ∗ φ is the set of ≺-

minimal states that are consistent with φ.

4 KNOWLEDGE

REPRESENTATION FOR

BUILDING SCIENCE

4.1 An Ontology for Building Science

Investigations

Looking at the case study, it is apparent that IAQ

investigations require a great deal of background

knowledge and expertise. It is also apparent that,

given the required background knowledge, solving

the problem involves enumerating all possible states

of the world that give rise the reported conditions.

One sensible solution, therefore, is to start with a

background knowledge base that consists of a large

set of constraints and dependencies between

residential conditions and health outcomes.

The first step towards the development of

practical tools to support this process may therefore

be the development of a formal ontology. This

essentially involves listing all of the relevant

components of a building, along with relationships

between them.

Knowledge-based Analysis of Residential Air Quality

803

Figure 2: Building Science Ontology.

This can be done through a knowledge acquisition

process with Building Science experts. One informal

ontology, originally presented in (Mora et al. 2011),

is depicted in Figure 2. Of course, this is not a formal

ontology; it is simply a partial diagram of key

building components. However, it would certainly be

straightforward to extend this ontology and translate

it into some variant of OWL for reasoning.

4.2 Reasoning about Static Building

Performance

Given the required background knowledge, IAQ

problem solving involves enumerating all possible

states of the world that give rise the reported

conditions. If we have a particular symptom, we can

then identify all of the minimal world models that

support the symptom in a non-circular manner while

respecting all background conditions. The natural

model for this kind of reasoning is answer set

programming.

The general approach being proposed here is to

develop a set of logic programs that encode expert

knowledge of building systems, where the answer

sets represent explanations of air quality problems. At

a very high-level, there may be simple propositional

rules, such as:

MoldInHouse



MoldInRoof

While other rules may state conditions on symptoms:

AggravatedBreathing



MoldInHouse, Asthma.

Significantly, these rules must span all relevant

background knowledge. For example, in the case

study described, every cause of aggravated breathing

must be encoded.

4.3 Reasoning with a Dynamic Building

Ontology

Ontology evolution occurs when a domain is

described by a formal ontology, and we acquire new

information about the domain that is not consistent

with the current specification. This occurs frequently

in Building Science. Suppose we start with an

assumption that there is mold in the roof. This will

have an impact on air quality, and cascading effects

on people living in the house. But two things could

change the way this impacts of knowledge:

1. We may look in the roof and discover there

is no mold after all.

2. We may remove the mold from the roof.

In both cases, changing our view on the mold in

the roof will impact our views on the building

performance. Formally, we have to change the

ontology representing our building.

Superficially, this is problem is similar belief

revision; many authors have proposed that the

methods developed in belief revision theory can be

applied to ontology evolution problems. However, as

noted previously, the most widely known approaches

to belief revision are propositional whereas ontology

languages such as OWL are variants of first-order

logic. As such, it is difficult in the general case to use

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

804

belief revision operators to capture ontology

evolution. However, in the concrete case of Building

Science, we suggest that this may not be a problem.

Our suggestion is the following. If the building is

being modelled by an OWL-RDF ontology, we can

divide the ontology into two components. The A-Box

can be translated directly into a propositional theory,

in which each statement about an individual is

translated into a propositional atom. The T-Box is not

translated into a set of logical statements; instead, the

T-Box is used to define a total pre-order over

interpretations for revision. Intuitively, the

plausibility of an interpretation is determined by how

many of the T-Box axioms are violated. This

approach is motivated by the fact that, in Building

Science applications, we actually do not want to

change the definitions of properties and concepts. The

theoretical advantage of this approach is that we do

not have to address any first-order issues in the

revision. The practical advantage is that it takes an

OWL-RDF input, and it can produce an OWL-RDF

output that differs minimally while respecting as

many conceptual axioms as possible. As a result, we

suggest that it would be possible to implement this

approach as a plug-in for the ontology reasoning tool

Protégé-OWL.

5 CONCLUSIONS

In this position paper, we have proposed formal

logical methods may be useful for reasoning about

building performance. At this point, it may appear

that the proposed solution is simply some form of

advanced expert system. There is a sense in which

this similarity is genuine: creating an ontology for

building science involves a large knowledge

acquisition effort in collaboration with domain

experts. This kind of interdisciplinary, practical

ontology development has already been effectively

carried out in other domains, such as medicine and

molecular biology. However, this practical effort is

not all that is required; the proposed solution actually

requires fundamental theoretical advances in

Knowledge Representation and Reasoning.

The main problem that must be addressed here is

the issue of ontology evolution. In many domains,

including building science, the basic ontology used to

represent the domain changes periodically as new

information is obtained. When the new information

conflicts with something in the ontology, some form

of conflict resolution must be employed to propagate

the new information throughout the ontology without

inconsistencies. Many solutions have been proposed

for this problem, based on existing work in other

areas of Computer Science such as Database Theory

or Belief Change. To date, however, there is not a

generally accepted approach to ontology evolution.

Our belief is that the concrete study of BSYS using

ontologies, answer sets, and belief revision operators

could provide a step in that direction.

It is also worth noting that using answer set

programming to reason with ontologies is an idea that

has previously been explored (

Magka 2013). As these

formalisms have been developed in parallel in

different AI communities, it has historically been

difficult to combine the two in a practical problem-

solving domain. Again, we suggest that this

application provides an appropriate domain for

reconciling these formalisms, while solving an

important practical problem.

REFERENCES

Alchourron, C., Gardenfors, P., & Makinson, D., 1985. On

the logic of theory change: Partial meet functions for

contraction and revision. Journal of Symbolic Logic, 50

(2), 510–530.

Baral, C., 2003. Knowledge Representation, Reasoning and

Declarative Problem Solving. Cambridge University

Press.

de Mast, J., 2011. The Tactical Use of Constraints and

Structure in Diagnostic Problem Solving. Omega

39(6), 702–709.

Morgan, W. J., Crain, E. F., Gruchalla, R. S., O'Connor,

G.T., Kattan, M., Evans III, R., Stout, J., Malindzak, G.,

Smartt, E., Plaut, M. and Walter, M., 2004. Results of a

home-based environmental intervention among urban

children with asthma. New England Journal of

Medicine, 351(11), pp.1068-1080.

Magka, D., Krötzsch, M. and Horrocks, I., 2013.

Computing Stable Models for Nonmonotonic

Existential Rules. In Proc. of the 23rd Int. Joint Conf.

on Artificial Intelligence (IJCAI).

Mora, R., Bitsuamlak, G. and Horvat, M., 2011. Integrated

Life-Cycle Design of Building Enclosures. Building and

the Environment 46(7), 1469-1479.

Mora, R., Croft D., 2013. Building Science Integrated

Systems – Methodological Framework, Architectural

Engineering Institute (AEI) Conference.

Motik, B., Cuenca Grau, B., Horrocks, I., Wu, Z., Fokoue,

A., and Lutz, C., 2012. OWL 2 Web Ontology Language

Profiles (Second Edition). W3C Recommendation. http:

//www.w3.org/TR/owl2-profiles/

Knowledge-based Analysis of Residential Air Quality

805