- From: <Michael.Compton@csiro.au>
- Date: Tue, 22 Dec 2009 22:49:29 +1100
- To: <Michael.Compton@csiro.au>, <public-xg-ssn@w3.org>
Hi all,
Kerry and I had a long discussion about this issue this afternoon and tried to assimilate all the various points. The three sections below contain our conclusions in text, one way of rendering it in DL, and open issues.
------------------
Text describing what we though...
A system contains components.
A process has an output and possibly inputs and, for a composite process, describes the temporal and dataflow dependencies and relationships amongst its parts.
A device is a physical unit / a piece of hardware / a system in a box.
Sensing is a process that results in the estimation of the value of a phenomenon.
An observer is a thing that can do sensing.
A sensor is a device that can do sensing
A sensing system is a non-device system that can do sensing (so this is intended to represent things like the regional observing system that Luis mentioned)
An AdHocObserver is all the other things that you might want to model that can do sensing (for example, me calculating wind chill - not really a system, but definitely doing sensing)
An observation is the record of some sensing and contains time stamps, a result, a reference to the sensing process ... i.e. all the bits in O&M
Notes:
There seemed to be some difference in what the word sensor mean (some people though devices, others thought of it as anything that can do sensing) so we decided to take sensor as devices and then categorise all the other things that can sense as some other type of observer.
Also, we agreed that coming to estimations of the value of phenomenon is a process, but both had difficulty with the idea that a device is a process. It seems more like a device can do these processes, but not really that a device is a process.
These definitions make a split between systems whose component relation describes structural make up, and processes whose component relation describes temporal ordering and data flow. SensorML seems to make a similar distinction by classing some processes as abstract things and systems as physical things (though it of course has a superclass of abstract processes that encompasses both).
------------------
An OWL representation of the key points...
I'll use <= for the OWL sub class relation and == for the equivalence relation
System <= hasComponent . System
Device <= System
Sensing <= Process
Observer == cando . Sensing
Observer == (Sensor or SensingSystem or AdHocObserver)
Sensor <= Device
Senser <= Observer
SensingSystem <= System and not Device
SensingSystem <= Observer
AdHocObserver <= Observer and not System
------------------
Open issues...
There is really one big issue in the discussion: how do processes relate to systems? We see 4 options
(1) Process <= System
(2) System <= Process
(3) state nothing
(4) Process disjoing System
Option (1) is essentially stating that the component relation for processes is a sub relation of hasComponent, which seems fair enough, but I'm not sure if it's essential or what it adds.
Option (2) is the SensorML modelling, but seems difficult for most people to come to terms with.
Option (3) allows either option (1) or (2), so is flexible, but it's not then clear how one could reconcile two ontologies where one uses option (1) and the other option (2).
Option (4) forces processes to be abstract descriptions of steps and temporal relations and systems to be structural relationships and the two to be different.
Michael
________________________________________
From: public-xg-ssn-request@w3.org [public-xg-ssn-request@w3.org] On Behalf Of Michael Compton [Michael.Compton@csiro.au]
Sent: Monday, 21 December 2009 2:32 PM
To: Semantic Sensor Network Incubator Group WG
Subject: [ExternalEmail] Re: ISSUE-2 (All processes are systems): All processes are systems [sensor ontology - http://mmisw.org/orr/#http://www.w3.org/2009/SSN-XG/Ontologies/SensorBasis.owl - 09.12.15 ]
Maybe I should just address the dot points.
- A process has inputs and outputs
ok
- A system has components
ok
(1)
- A sensor is a process
(2)
- Some devices are sensors
- Some devices are sensors - because they have an output
are we saying that a sensor is a process or that sensing is a process. For example, lets say:
* I have a device that measures wind chill (i.e. it measures temperature and wind speed and does a calculation), by (2) above it's a device that is-a sensor, and by (1) it must be a process (I'm uncomfortable with this already because we have now said that a device is-a process, which seems wrong to me. I'd think a device might follow some process, but this is-a relationship seems strange).
* Now what if I write down the wind speed and temperature measurements myself and do the calculation myself. What's the sensor here? It can't really be me, can it - it would seem strange to say that I am a process and a sensor. Seems more like I followed a process and thus calculated wind chill. So maybe the sensor is the process I followed? Or is it the act of me following the process? In either case we have a problem because above we said a device is-a sensor and then here we are saying something entirely different (a process or the act of following it) is-a sensor.
I would say that a device cannot be a sensor (well not in the process sense that we have been talking about) otherwise we are conflating an abstract (a process) and a concrete thing (a device).
Seems from all the discussions that we have had that sensing is-a process - or that some processes result in sensing something, and that a device or a person, or a regional observing system might act out such a process and thus sense something.
So I would be more comfortable with
- Sensing is a process
- Some devices can act as sensors
And then that a device that senses something could be a 'sensing device', which thus acts out some sensing process.
but I don't agree with (2) above
- Some devices are systems
Why aren't all devices systems? Even if they only have one component or we don't want to write down all their components?
- Some systems are sensors
It depends, is our definition "A system has components" or "All things with components are systems"? The question itself is silly, but my point is why are we trying to use the same component relationship to describe devices and processes?
I'm having trouble seeing the example above with me calculating wind chill as a system.
Michael
On 20/12/2009, at 2:19 , Luis Bermudez wrote:
Hi Michael,
The only thing we have said about systems is that it contains components...
On Fri, Dec 18, 2009 at 12:42 AM, Michael Compton <Michael.Compton@csiro.au<mailto:Michael.Compton@csiro.au>> wrote:
my conceptions/preconceptions/misconceptions are as follows.
- A sensor need not be a physical device. Kevin's definition of "An entity capable of observing a phenomenon and returning an observed value." seems ok to me.
agree !
- A Sensor need not be a single entity - it can be composed of any number of sub sensors, each perhaps of their own identity, each perhaps measuring different things that get combined into the whole.
This is why it can be a system . Maybe we need to add also:
Some sensors are systems
- The following things 'are' sensors
*a temperature sensor (i.e. a physical device) at location l
*a program on a computer (far away from location l) that can read in the temperature at location l and a wind speed estimate for location l and calculate the windchill at l
Yes.. Commonality is that a sensor has an output and therefore are processes.
More correctly, in both cases something has acted as a sensor for a particular phenomenon: a device in the first instance, and the program in the second - if I wrote down the temperature and wind speed measurements on a piece of paper and calculated the wind chill myself, then I have acted as the sensor.
- The example of the wind chill sensor means that sensors can have multiple components, and I guess the components may themselves be interesting.
- A device (a physical piece of hardware) can also be broken down into components (presumably other devices, but perhaps also systems - software systems etc) but I don't see that as having anything to do with sensors or their decomposition into parts. For example, imagine a device that can measure wind speed and temperature, that has a small inbuilt chip that can calculate wind chill, round measurements, compute averages and a radio to communicate its readings. It's physical decomposition into its components is different from its decomposition into the roles it can play as a sensor. So the two sorts of decomposition may be related, but not equivalent.
So I think this still holds true:
- Some devices are sensors - because they have an output
- Some devices are systems - because they have components
But if you wnat to propose other statements which make the system composition more explict, please do so.
- I'm also unsure about the word system and in particular it's relationship to process.
We are separating them..
- partly I see the problem as linguistic: i.e. we are using the word sensor in two different contexts. We think of things in terms of 'ah this thing is a sensor', but we also say 'a sensor is a process'. Is what we are really saying that to sense something is to follow a process that leads to a value as an estimate of a phenomenon. In which case a sensor isn't a thing at all it's really a 'to sense do...' or 'was sensed by doing...'. So if we take it that to sense something is to follow a process that estimates a value, then what is a system and why is a sensor one? To think of a system as a collection of components in some technical sense and then make sensor one of these is to take the 'ah this thing is a sensor' approach, but then we also agree on 'a sensor is a process' which now seems to make a sensor not a system. So is the biggest problem here simply that we (copying from SWE) have decided that systems have components and sensors are also made up of things, so it must be a system - where as there are actually two hierarchies here and we should represent them with different relationships?
The only think we are saying about sensor is that it has an output !
So how about....
a System is a device/computer system/software system that is made up of components
I think we do not need to be explict about device/computer system/software. For example, a regional observing system can also be a system.
a Sensor is a process (a description of inputs/outputs, some steps and data flows) which may also be made up of sub sensors
yes.
a System may play the role of a sensor for phenomenon X.
So is this OK ?
- Some systems are sensors
but I suggested before that Some sensors are systems
I think both are ok.. what do you think ?
-luis
Michael
On 17/12/2009, at 19:11 , John Graybeal wrote:
On Dec 16, 2009, at 12:00, Manfred Hauswirth wrote:
Hi John,
thanks for your insightful comments. Some more comments from my side.
John Graybeal wrote:
Regarding "all systems are processes": Honestly, I would not >
understand this (I stated this at the F2F). For me, you have
systems > which include one ore more processes. If systems are
processes, why > have systems at all. My notion of systems would
informally consist > of processes, scenarios, deployments, etc.
The question "why have systems at all?" is the crux here. Can we
state clearly when a process is not a system? Or in other words,
how is a system more narrow than a process?
Incidentally, my notion of processes would informally consist of
the same list. I am also having trouble drawing the distinction.
Interesting! I think this may be due to our different background (I
assume your are not a computer scientist like myself - without
evidence I may add).
Computer Science and Statistics. 30 years software and systems
support. (No worries!)
In my area (computer science, information systems) systems would be
defined as I do and a system would consist of software and hardware
and the processes would clearly be "inside" the system as part of
the software, so there is a clear distinction between "system" and
"process" (other CS/IS people - please feel free to contradict me),
whereas you seem to define this more from the viewpoint of an
experiment which is being observed (?) where processes come into
play as part of the observation process (please correct me - I am
guessing here).
I'm using one of the general meanings of the word 'process', which
applies not just to what's happening in side the computer or
component, but what happens as all the software and components
interact with each other. There are local processes and there are
external processes.
It isn't driven by experiment orientation but by broader CS
orientation -- dealing with engineering systems of systems, and
including the human component in those systems, and modeling all the
above as processes (which may, or may not, then be computerized in the
new version of the system). Anyway, just a different viewpoint,
neither right nor wrong.
The problem here seems to lie in different conceptualizations in
different communities - all of which done according to the specific
needs of a community. Now, while this may complicate things, I think
it is also a useful and actually mandatory exercise. While I may
claim, that I need to understand the conceptualization because as an
CS/IS person I will have to build (software/hardware) systems
(sorry! no other term comes to mind) which need to manage
information coming out of observations, you may claim exactly the
same from you point of view (and rightfully so). The question now
for me is: Who are our users and how to serve them best? Where's the
sweet spot?
Concur. I presumed from the start that the group was interested in
modeling hardware elements, but I have found it useful to consider
those hardware components as processes in a larger system of systems.
They take data in and transform it to other data that is spit out.
This is one useful definition of a process, as Luis notes.
Oops, got off track there! But our agreed point is to agree on which
type of devices (= which group of users) we want to make the ontology
for. My assumption/preference was the group that used physical
devices to transform measurable phenomena into digital data (because
that's the easiest to model and the most immediately useful). But I
can go with whatever on this, as long as we all understand.
PhysicalSystem: I don't remember the exact reason for this. Did
we > mean deployment?
I assume this is to distinguish it from a software system.
Sensor as subclass of Device: I think this is too narrow. I can
think of sensors which are not devices at all, e.g., human
"sensors" > in the context of social sensing (which is an accepted
concept in > many domains including CS by now). Making sensors a
subclass of > device limits us to purely technical systems in
hardware, IMHO. Is > an RSS feed a device? I can clearly use it as
a sensor. I think that > Device should be a subclass of Sensor.
Even in existing middelware > systems like our GSN we followed
that path (without having an > ontology in mind at all).
This gets to purpose of the ontology. As I understood it, the
group was originally constructed to model hardware sensors. (May
have just been a wrong assumption on my part. More precisely, what
we clearly were not doing is modeling samplers, that is, devices
that return a physical sample.)
Agreed. But "sensors" do not necessarily manifest themselves as
hardware. If I want to detect user activity / inactivity on a
computer in an experiment, one of my sensors may be a the keyboard,
another one running processes (not waiting for user input), etc. It
is very hard to draw the line here. My question: Do I have to have
this distinction at all? Essentially I convert an X into a Y and Y
should be usable in a computer. Whether X a is a physical phenomenon
or not depends on the domain, IMHO.
Sure, that works for me too. If you make a sensor too general,
though, it can have components. What do we call those components --
are not at least some of them sensors? So now, what is different from
the sensor that can have sensors, and a device, which has the same
recursion into smaller devices; and a system, which can have systems
(and a process, that can have processes)?
I'm being a little silly of course. All I mean to do is call
attention to the need to define the terms according to what makes them
different from each other, not just whether they are higher or lower
in a hierarchy. I think we haven't done that well enough yet.
So using one definition of sensor ("anything that senses") makes
Sensor very broad, and other things would subclass to it. (Since
some devices (a hammer) don't sense things, we'll have to define
Device narrowly to make it a subclass Sensor.) Using another
definition of sensor ("a component that detects (measures) a
physical phenomenon, converting it into a digital representation
that can be output to other components"), a Sensor is clearly a
specific type of Device, and is also a component of any sensing
device.
If you see software as a Device, I would agree to it, but then again
Device has the connotation of hardware.
Ah, I said a Sensor was hardware in my original world, so I didn't
have any problem here -- since my Sensor was hardware and my Device
had a sensor, I was already on board with Device being hardware.
Do we have a set of definitions by any chance, so we can all use
these (or some) terms the same way?
I don't think we have.
Why is a Device a subclass of a Process? A Process can use
Sensors > which are manifested as Devices to do/measure something,
IMHO. Again > this is a quite narrow notion of the concepts.
I'm not following your argument here. Yes, a Process can use
Sensors as you say. So can a Device. There is no inconsistency
that I can see. This suggests a Device is in fact a type of Process.
Sorry, but I don't understand how a Device can be a Process.
The "Process: something that receives an input and produces an output"
is not a sufficient explanation or model of that?
John
Best regards,
Manfred
--
Luis Bermudez Ph.D.
Coastal Research Technical Manager
Southeastern Universities Research Association (SURA)
bermudez@sura.org<mailto:bermudez@sura.org> - Office: (202) 408-8211
1201 New York Ave. NW Suite 430, Washington DC 20005
Received on Tuesday, 22 December 2009 11:51:03 UTC