We are seeing the emergence of the Internet of Things, where digitally enabled objects can interact in smart environments. RFID can play an important role in linking common objects to the digital world. In this paper, we focus on e cient processing of collective waste items. These are considered to be smart by tagging them with RFID which bears the description of its properties. We have demonstrated a model using OWL ontology to sort these smart waste items for better recycling of materials. Our motive for using ontologies is for representing and reasoning of the domain knowledge to be autonomous, reducing the need for frequent references and updates for knowledge de nitions from external sources in real time.
Today Pervasive computing is gradually entering people's everyday life.
Commonly used objects are made smarter. They are able to adapt and integrate
with the environment which might have capabilities to perform computation
and information processing. The use of RFIDs and sensors have an important
contribution in imparting intelligence. RFID chips are generally used to connect
physical objects in the digital environment. Presently, RFID tags are widely
used by retailers and manufacturers for inventory management. We can assume
that in future they will tag every product. This would also support for better
waste management and recycling. Considering this future trend, we consider to
have \self describing" smart waste items. E orts have been made in the past to
perform e cient waste management using RFIDs [
The generation of waste items in recent times has been growing by leaps
and bounds [
Every category of the selective sorting process would have some criteria and
conditions like the amount and quality of recyclable materials recovered. In addition
to this there may be some hazardous or unwanted waste items that might cause
contamination to this category [
Taking examples from everyday life, some of the recyclable materials are di erent types of glass, paper/cardboard, plastic, metal etc. We would take examples for our model that elucidate the selective sorting of some of these materials. Hence our ontology contains classes such as Glass, Paper, Plastic and Metal. Being distinct types of materials, they are represented as disjoint classes in the ontology. They represent the generics and they contain individuals as members to represent speci c categories. Speci c types of glass like SodaLimeGlass, FoamGlass etc. are listed as indivduals in the ontology. Our model that accepts or rejects items depending on the preconditions that are set regarding the categories and amount of recyclable materials it wants to maximize. 2.2
The W3C group provides the ontology design method for N-ary relations in an
ontology [
Formally, an N-ary relation forms a mapping of an individual with other individuals or data values. The relations are all interconnected in some way. It is used when additional information need to be added to a binary relation.
The W3C lists four types of cases where N-ary relations are useful. The rst case demonstrates the use of N-ary relations when additional attributes need to be provided with a relation. The classical example they have stated is \Christine has breast tumor with high probability".
The N-ary relations referred above have provisions for specifying the probabilities for the diagnosis as discrete values. This use case suggested by W3C could be modi ed to represent some speci c situations. The attribute that relates the probability on the diagnosis consists of a discretized value. It would be more realistic if the probability values could be represented with numeric values. Hence we have incorporated this along with some more changes to build a model that exhibits useful features. We will show the usefulness of this model in practical applications which would be discussed in subsequent sections.
Modifying the rst use case of W3C, we have de ned the N-ary relations linking the classes as shown in gure 1. The class Diagnosis Relation 20 interconnects all the related classes we are interested to form relationships with. Its relation with the class Disease using the property diagnosis value expresses a diagnosis with at least one disease. Similarly the other property diagnosis probability holds all relations with integers having any value 20. This is one of the modi cations we had stated earlier which makes the numeric representation of the probability value more realistic. We have de ned both diagnosis value and diagnosis probability as functional properties, thus requiring that each instance of Diagnosis Relation 20 has exactly one value each for Disease and its probability.
Figure 1 has another property has diagnosis that represents relation between the class SickPerson and Diagnosis Relation 20. The existential quanti er \someValuesFrom" requires to have at least one instance of the class Diagnosis Relation 20. So this completes the entire relation to represent a SickPerson diagnosed with a disease with probability 20.
Now we will de ne another similar N-ary relation class Diagnosis Relation 40 having probability value 40 and make both these classes equivalentClass. We are putting down the complete de nitions for these two classes below:
We would observe an interesting feature with these classes. The class Diagnosis Relation 20 would be inferred as superclass of Diagnosis Relation 40 by the reasoner. Hence the individuals in the class Diagnosis Relation 40 satis es both the class conditions. Figure 2 represents the sets of classes using a Venn diagram. Reiterating on our problem described earlier, we intend to perform the reasoning of this selective sorting based on the percentage of recyclable materials it contains as well as maximize the extraction of the valuable recyclable materials? In this section we will explain how our implemented model performs this speci c task. This model makes extensive use of the modi ed N-ary relation using numeric values as explained in section 2.3.
The system performing selective recovery should have su cient knowledge
for making decisions to accept or reject a waste item. We have chosen to embed
this knowledge using ontologies in the systems. Ontologies can formally represent
a set of concepts within a domain along with the relationships between them [
Before proceeding with description of the various categories using ontologies, we would introduce the used N-ary relations. An N-ary relation such as Glass 30 in gure 3 would represent Glass with quantity more than 30%. There are two properties material value and material percentage that links Glass 30 with the recyclable material class Glass and the numeric value 30 as percentage. Plastic 50, Glass 80, Metal 60 and Paper 70 are the other similar relations used in our ontology. We have used consistent naming convention for all such N-ary relations in our ontology. Electronic waste items or simply e-wastes constitutes any type of electronic item like computers, telephones, televisions and so on. These contain various types of materials both useful as well as hazardous. Materials like plastic, glass etc are contained in them which can be used for recycling. To identify and perform selective sorting of these types of waste items using the ontology we have de ned an equivalent class eWasteBin that considers any item as e-waste which contains more than 60% of plastic combined with more than 30% of glass. Capturing these conditions, the axioms can be stated using Manchester syntax as: This item would be classi ed as an electronic waste item satisfying the conditions of eWasteBin.
What happens if we are interested in sorting of items that contains only one major type of recyclable material. For instance, GlassBins would like collect items selectively that contains Glass of more than 80%. The following class represented below would be able to capture such types of items:
For the representation of this category there exists one and only one N-ary relation. So we have expressed the same using both the OWL existential restrictions allValuesFrom and someValuesFrom.
A typical example containing high amount of Glass would be coke bottle. The axioms for such an individual would be:
There is a subtle addition while stating the properties of the individual CokeBottle. Due to the open world assumptions of OWL ontology, it needs to be stated explicitly to contain only one type of material. So the axiom \contains exactly 1 Rel Material Qty" is added to assure the OWL reasoner that this individual contains only one type of material. As we are using the model and add items through automatically generated programs, it would not be di cult ensuring to close o this information. Classes for other categories can be created similarly to the GlassBin. Figure 4 shows the visual representations of MetalBin and PaperBin in the ontology. They use the modi ed N-ary classes Metal 50 and Paper 70 respectively.
These various categories can also be described by class expressions without using the N-ary classes such as Metal 50, Paper 70 etc. The quantities for the recyclable materials are standardized for each category and won't be quite alot in number. So using these named classes would make the description of categories simple and easy to understand. We can also reuse them for de ning newer categories as well as utilize them in DL queries. 4
Until this point, we have discussed our model for representing knowledge for
selective sorting of collective waste items using ontology. The discussion also
includes ways to represent items as individuals to be classi ed in the proper
category. Our application has used the Information model in the backend.
Ontologies are increasingly being used in Pervasive Computing Environments as
some of their features are very well suited for the purpose [
The smart waste items are tagged with RFID tags that contain the details
about the percentage contents of the various recyclable materials. For example,
the individual ItemX in section 3.1 would have the data containing in its tag as
Glass-35,Plastic-62 as plain text. Our intelligent bin would have a RFID reader
to detect the smart waste items and read the information contained in its tag.
The application running on the bin would use the ontology based knowledge to
infer the item's category. If found suitable, the bin would accept the item. The
ontology based model can also be used to sort items in waste processing plants.
In our project Bin That Thinks, we are developing such applications using the
discussed model in consultation with a major waste management company [
We can use an alternative method for tagging the generic smart waste items. For example, a generic item like Monitor can have an entry in the ontology as below:
Hence while tagging monitors, it would be su cient to write this classname Monitor into the tag. Our application would be able to detect such tags for proper categorization. 5
In this paper, we have proposed a model that can classify waste items based on
the recyclable materials they are made up of. Our model utilizes the concept
of N-ary relations with some modi cations. It is very useful in the context of
selective sorting. Waste processing plants have automated selective recovery
systems to maximize the collection of recyclable materials with least contamination
of hazardous or unwanted waste items. Currently, our ongoing work proposes
the composition of recyclable materials to be indicated on the RFID tags
afxed on the items [
In our future work we propose to utilize the full bene ts of ontology by performing the domain knowledge sharing among the autonomous sorting systems in the entire infrastructure.
Acknowledgments. We would like to thank the Protege user community for valuable suggestions while developing the ontology model.