What is Meant by "Level of Accuracy?"

A 2.346Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

As a service provider who documents buildings I often hear my clients say "I need one quarter inch accuracy" or "everything , needs to be within an eighth of an inch" or, some other stated level of accuracy. How can I be certain I am able to provide what my contract requires when that’s all that the specification states? How much risk am I assuming by proceeding with the work without a better definition of what the stated level of accuracy means? It seems like these statements are made and they are agreed to, but no one is really sure what they mean or how to measure whether the specification was met, especially when it comes to the process of scan-to-BIM.

Specifying level of accuracy is something that the AECO (Architecture, Engineering, Construction and Owner/ Operator) industry has long struggled with in regards to existing conditions building documentation. What is meant by accuracy? Is accuracy relative or absolute? These are things that, if left undefined, can put both clients and service providers at risk with regards to the as-built deliverables.

A common practice in the design industry is to represent 2D or 3D building data in an idealized, orthogonal graphical view. Real world existing conditions are seldom orthogonal. The challenge then becomes how does one represent the real world, non-orthogonal conditions? Are they to be represented orthogonally or with respect to their true out-of-plumb conditions?

To complicate matters further, many of today’s design software packages prefer to work in an orthogonal graphics environment and are limited in their ability to accurately represent out-of-plumb conditions. When real world, out-ofplumb conditions are represented in an orthogonal fashion error is introduced. When error is introduced and the specification calls for a specific level of accuracy how does this get reconciled?

In addition, when documenting existing conditions it is common to encounter hidden or concealed conditions making it difficult or impossible to document certain Levels of Development (LOD) and/ or achieve certain Levels of Accuracy (LOA). When these conditions occur, what are the expectations and what can actually be achieved with regards to meeting the required accuracies?

Finally, it is important to recognize that the measured level of accuracy acquired through data capture may be different than the level of accuracy represented in the processed data. Each of these things needs to be carefully considered when specifying a level of accuracy. So what can be done to ensure required accuracies are met and risks are reduced?

Specifying a Level of Accuracy
Let’s start by asking what is meant by the term accuracy? Often people get confused with the difference between accuracy and precision. To illustrate, Figure 1 shows a series of targets with individual points (or measurements) colored in grey. The mean value of the individual points is shown by a red colored point. The objective is to measure to the center of the target with the highest degree of accuracy. The center of the target represents the true position of the point on the object we are trying to measure (also referred to as the standard value). Figure 1B illustrates that we can be very precise in our measurements without being accurate. However, measurements can also be considered accurate without being precise as seen in Figure 1A. Finally, we can think of the definition of accuracy as "the extent to which a given measurement agrees with the standard value for that measurement."

Another important aspect of specifying accuracy is defining whether the acceptable level of error is absolute or relative.

Absolute error is the amount of physical error in a measurement, period. Let’s say a tape measure is used to measure the width of a room. The acceptable level of error is inch +/-. This is the absolute error of the measurement.

absolute error = inch (0.02083 ft).
or
absolute error = x

Relative error gives an indication of how good a measurement is relative to the size of the item being measured. Consider two surveyors are measuring a building. The first surveyor measures the width of the room with a tape measure and the second surveyor measures the length of the building with a total station laser. The first surveyor gets a value of 10 ft. with an error up to inch (0.02083 ft.). The second surveyor measures the length of the building resulting in a measurement of 100 feet with an error up to inch (0.02083 ft.). Upon investigation we can see that the overall accuracy of the 100 ft. "building length" measurement is much better than that of the 10 foot "room width" measurement. The comparative accuracy of these measurements can be determined by looking at their relative errors.

relative error = absolute error value of thing measured
or
relative error = delta x x

where x is any variable. Now in our example,

room width relative error =

0.02083 ft 10 ft 100 = 0.2083%

room length relative error =

0.02083 ft 100 ft 100 = 0.02083%

Clearly, the relative error in the building length measurement is considerably smaller than the relative error in the room width measurement even though the amount of absolute error is the same in each case.

Orthogonal vs. Real World
As an architect, my first lesson in representing existing conditions orthogonally vs. in a real world fashion came when I first started creating building as-builts for other architects. I was always taught to draw plans in an orthogonal fashion. Before CAD, we used tools like T-squares, triangles and parallel bars to create our drawings. When CAD came along the only thing that changed was our orthogonal lines got straighter and more precise. Our method of representing walls and other building elements remained the same.

When I measured a building with my tape measure and clip board I would always round the fractions to something easy to work with and draw my lines parallel and perpendicular to each other. Of course the building never closed properly because of the error I was introducing through my process and the low tech accuracies of my instruments.

When I obtained my first hand held laser range finder, total station and CAD enabled field computer I decided that I could now improve my process by eliminating the rounding error since my instruments were so precise. I stopped drawing my lines in an orthogonal fashion and I found my accuracies improved. I was able to obtain less overall error and my buildings closed better.

However, I soon found that my clients were rejecting my deliverables because they were not drawn orthogonally. They didn’t like to see or deal with pixel jogs on their screens. Ironically, because my drawings were not drawn orthogonally my clients began to lose confidence in their accuracy. I quickly developed a new field process that allowed me to draw orthogonally, yet minimize my error to what my clients and I considered to be an acceptable architectural tolerance. My clients were happy once again.

Now fast forward to the advent of laser scanners and BIM (Building Information Modeling). Laser scanners are survey grade instruments that can quickly capture millions of measurements and represent out-of-plumb real world conditions with extreme accuracy. These measurements are presented in the form of a point cloud. Not too long ago some BIM software platforms developed the ability to import and work with point cloud data. This was a significant game changer for the industry and improved my ability as a service provider to quickly capture and represent existing conditions.

However, with new technologies come new challenges. The BIM software prefers to work in an orthogonal fashion. Wall and pipe joins do not fit properly and make watertight connections unless they are joined at right angles. However, if you join them at right angles then the modeled element won’t necessarily follow the point cloud data (see Figure 2). When modeling point cloud data of "real world" pipes with sags, or walls and columns that are out-of-plumb and skewed (see Figure 3), how does a modeler reconcile these seemingly opposing forces and still stay within spec?

Many clients do not fully understand this, nor is it something you can explain in a 30 second elevator speech. In fact, I’ve found that clients often don’t have the desire or patience to be educated. Now, add that quarter inch accuracy spec and you may start to wonder how you are going to fulfill your contracted obligation.

Exposed vs. Concealed Conditions
When creating a scan-to-BIM deliverable I often use a phrase I call "a geometric BIM." Before I explain what this means, it is important to recognize one of the key advantages of using a BIM vs. a CAD model. BIM is a parametric modeling software. In other words, the elements within a BIM can be constructed by defining parameter variables in lieu of using vector line work. This is, in part, what gives a BIM its intelligence. These "smart" elements can then be used to perform various complex analyses.

Because a laser scanner only captures and represents point data, there is no intelligence associated with the acquired data. Meaning, a series of points representing a wall plane are just a series of points representing a wall plane. There is nothing to indicate whether these points belong to a wall, a door, or a piece of equipment. In addition, since lasers do not see through or into objects, they can only capture the surface of what they scan across. In a sense, they are only capturing the geometry of the objects they scan.

Since the laser only captures the geometry of exposed object surfaces, it is not possible to know or represent what is inside of those objects without other investigative techniques. Examples of such items might include a fireproofed steel member, an insulated pipe, or a wall assembly. Without knowing what’s inside these elements/assemblies we can’t accurately represent them in BIM as a smart object. Instead, a common approach is to model these objects to the point cloud as generic elements or modeled-in-place elements. When this is done the focus is switched from creation of a fully parametric intelligent BIM to a "geometric BIM" or a BIM that shows , where things are, but not necessarily what they represent.

In the case of a fireproofed steel member other decisions need to be made. For instance, often a contractor is concerned with routing items past a fireproofed steel member without having to scrape off the fireproofing. Since the laser captures the outside surface of the fireproofing it would make sense to model to this surface. However, BIM will have a difficult time representing an uneven surface such as this. Often a visual best fit approach is used. Which poses the question: in such cases, is the service provider still required to meet the accuracy specification? What if the client insists on the service provider maintaining the accuracy spec for conditions such as these?

Accuracy of Acquired Data vs Modeled Data
The Level of Accuracy attainable is closely related to the tools, techniques and processes used to capture and represent the objects being documented. In the case of laser scanning, each scanner has a stated collection accuracy. The methods and techniques used by service providers to capture scan data such as overlap, point density, noise, point spray, methods of registration and control, all affect the accuracy of the registered point cloud data. Assuming that the acquired scan data can meet the accuracy specification, what happens when the provider is required to model that data? Doesn’t the modeling process introduce more error and inaccuracy?

We have already reviewed some of the challenges of maintaining accuracy during the modeling phase of a project. Modeling scan data is an interpretive process. Often point cloud data contains shadow areas, is not complete, and/or is difficult to process even by highly experienced modelers. Best fit approaches are often used to reconcile the differences between the real world conditions and the BIM software. At the end of the modeling process the modeled data will not be as accurate as the data from which it was derived. Unless this is understood and accounted for in the level of accuracy specification and contract, both the client and the service provider can be putting themselves at risk.

John M. Russo, AIA is an experienced architect with more than 30 years of experience in performing building documentation. in 1997 he founded irvine, California based, Architectural resource Consultants (ArC), www. arc-corporate.com , a firm that specializes in documenting existing building conditions. in 2012 he founded the U.S. institute of Building documentation (USiBd), www.usibd.org.

A 2.346Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE