How did automated specification inspection software develop?

author | Brian Potter       translator | Meniscus

Produce | CSDN（ID：CSDNnews）

The concept of automated specification checking comes from “ Automated building code inspection ”—— Software to check whether the architectural design conforms to the specifications . Many of the ideas mentioned in this article have been tried by the construction industry a long time ago .

below , Let's take a look at the development history of automated building code inspection system , And what lessons we can learn .

The basic idea of automatic standard inspection

First , Let's explain the background knowledge . New buildings must meet the requirements of local building codes , This series of rules covers all aspects of buildings , For example, the firmness of concrete 、 Height of handrail 、 The distance between staircases must be shorter than X Ruler, etc . The volume of the building code is very large , Almost every jurisdiction in the United States is using 《 International Building Code 》, There are only basic specifications 700 Multi page , And it just outlines a series of design requirements , Specific details need to refer to hundreds of documents and Design Standards . According to the level of in-depth reference , All these requirements may add up to 10 More than 10000 pages , But a building is usually only subject to some of these requirements .

The government is very concerned about whether the buildings comply with the specifications , So in the process of applying for building permit , Plan reviewers from the licensing office are required （ Sometimes there are multiple plan reviewers ） Review the design of the building , To ensure that it meets the specifications . The whole review process is very time-consuming , For example, in King County, Washington , At present, it takes time on average to obtain a building permit 20～30 Zhou . And for San Juan in Washington , The average planned review time is 12～20 Zhou . These are for single family housing , They are small buildings , The building codes that need to be followed are 《 International Housing Law 》 It's simpler . If it is a commercial building , Things tend to get worse .

Because the plan reviewer always puts forward some opinions that the designer must solve , such as “ The slope here does not meet the drainage requirements ”,“ There are too many glasses on this surface / too little ” etc. , The review time is bound to be extended , The design team needs to repeatedly pull with the competent department for several rounds , To get permission . Partly because of the complexity of building codes , It makes it easy for designers to ignore some of these regulations （ In particular, there are very different provisions between different jurisdictions ）. This is partly due to differences and inconsistencies between plan reviewers and jurisdictions , Their interpretations of norms are also different . for example , A study found , Under the same regulations , Violations of the same building in a jurisdiction are 0, In another jurisdiction 16 Breaking the rules . Plus, the United States has 2 More than 10000 licensed jurisdictions , You need to spend a lot of time to determine whether the design of the building conforms to the specifications .

This coordination process will lead to increased costs （ Because designers and jurisdictions 、 The contractor and the inspector need to talk back and forth about whether they meet the requirements of the specification ）, In the worst case, it will also lead to major risks in the project （ Jurisdictions do not accept certain systems , Or the acceptable modification scheme is completely infeasible ）. therefore , Every construction project will spend a lot of time and energy to determine whether it meets the requirements of the specification , And huge costs , For new or innovative building systems , This expense is particularly staggering . for example ,2021 In the U.S. 《 International Building Code 》 The regulations on large timber buildings have been added , However, most jurisdictions have not yet adopted , And even jurisdictions often do not know the jurisdiction 、 What does the fire management department accept , Don't accept anything .

In order to open this crux , One way is to use an automated specification inspection system , That is to analyze the architectural design through software program and automatically determine whether it meets the requirements of the specification . In an ideal situation , People trust this software enough , And accept the inspection results as proof of compliance , Just as the vehicle registration department accepts the emission test to prove that your car meets the requirements of vehicle emission regulations .

If you can really reach this level of acceptance , The licensing process in the United States can be greatly simplified . Not only can the judicial authorities reduce the time and energy of reviewing building permit applications , And architects and engineers can also be sure that the design they submit meets the requirements of the specification , So as to avoid back and forth with the licensing office .

As jurisdictions accept software inspections , Such software may become an abstraction layer for thousands of licensing jurisdictions in the United States , Unify and simplify licensing requirements .

This is a very attractive case , So since architectural design began to rely on computers , People have been trying to develop an automated specification inspection system .

A brief history of the development of automated standardized inspection system

We can roughly divide the automatic specification inspection system into three problems that need to be solved ：

●  Express building code requirements in machine readable form ;

●  Represent the building itself in machine readable form ;

●  Let stakeholders accept that the output of the software conforms to the specification compliance （ The output of software must be reliable 、 Verifiable and readable ）.

（ here , We can also add the fourth point , That is, perform the actual verification , Although most people think that as long as the above three problems can be solved , The fourth point is not difficult .）

Because automated specification checking has tried to solve different parts of the problem , Now let's look at each sub problem separately （ In fact, the distinction between these jobs is very vague , There is a lot of overlap between them .）

Express the specification requirements through software

In the early , The focus of automated specification inspection is to create software that represents building requirements , from 20 century 60 s ,Steve Fenves We are committed to transforming architectural design codes into large-scale decision tables , This novel representation of decision-making process is not only convenient for human understanding , And easy to build into software . As early as 1969 year , He and the American Steel Structure Association （American Institute of Steel Construction,AISC） The decision-making table of steel specification was worked out in cooperation ：

The focus of this work is not only to study how to check the steel specification through computer programming , As early as 20 century 60 years , The program written by the engineer can complete the engineering calculation , And compare it with the specification requirements . this , Their goal is to establish a reliable process ： First , Create a canonical format that can easily evaluate logical consistency （ Decision table ）; then , Implement the specification , And ensure that the final output can be trusted . The practice at that time （ It's the same now ） Is to standardize “ Hard encoding ” Into the software , This makes its output difficult to evaluate . Specifications are buried deep in the pile of software if-else in , Unable to verify whether these implementations are correct .Fenves Imagine a more robust process , Realize the specification requirements through an independent data structure （ This makes it clearer , And convenient for inspection ）, Then by a “ Request processing engine ” To deal with it .

throughout 20 century 80 years ,Fenves And others have made great efforts in this field . He and the National Bureau of Standards （ Now is NIST） Jointly developed SASE System , This is a way to create standards , It can ensure that the generated standards can be interpreted by the machine . later , He and AISC cooperation , Develop electronic version LRFD standard . Besides , He also helped develop SPEX engine （ This is a kind of “ Standard processing engine ”, It can be implemented independently of other specific standards ）.

besides , Others have also introduced similar standard processing systems , for example SICAD System , This is also a standard processing engine , National Highway and Transportation Association Standards （AASHTO） This engine is used . However , Except for a few projects that have made some achievements , Most of the work in this field is still at the academic level .

20 century 90 years , Building information model （BIM） The development of has opened up a new situation for solving this problem .BIM Allow computer modeling , Represent buildings in the form of large databases , As this modeling becomes more and more popular , People began to develop “BIM Check the engine ”, This kind of software can obtain a series of requirements （ These requirements can be built into the software , It can also be programmed later ）, And check BIM Whether the model meets these requirements .1996 year , Finland has developed Solibri, Later, this software became the most widely used software in this field （ Still in use today ）. besides , also FORNAX and Express Data Manager And other systems . In theory , It is not too difficult to automatically check whether the design of the building conforms to the specifications , We just need to provide the appropriate rule list for the inspection engine , And many automated specification inspection systems will also adopt this method in the future .

Singapore government Corenet The project is the most ambitious of these systems , The goal of this project is to build an automatic standard inspection system , Of course, there are other functions . Designers can upload the electronic version of architectural design , Then the software automatically checks its compliance . The first iteration of the software was 90 Began to develop in the s （ The plan is as early as 1982 It was proposed in , Even before BIM）, The initial plan was to use two-dimensional objects CAD. Later, a BIM The system has replaced （ The system uses FORNAX Check the engine ）.

Although the project ultimately failed to succeed （ Like many failed projects, he died quietly , Unknown cause ）, But lately Corenet-X The project is revived . The objectives of the project remain unchanged , But this time they used Solibri Model checker .

There have been similar projects in the United States , such as SMARTcodes and AutoCodes etc. .

SMARTcodes yes ICC stay 21 An ambitious attempt at the beginning of the century , Their goal is to develop such an automated specification inspection system .ICC The original plan was to develop software that could interpret specifications , And then through BIM Model checking software for evaluation .

because ICC Is essentially the creator of all building codes in the United States , Therefore, allowing jurisdictions to adopt this system can greatly simplify the process . Although the project has made some progress （ They even developed software to transform specifications into model checking rules ）, But they lost their financial support after the financial crisis , The project also died prematurely .

chart ：SMARTcodes Rule creation tool used

AutoCodes It is also a similar project , from Fiatech（“ Capital project stakeholders ” Industry Alliance ） On 2012 Launched in , yes “ Reproducible building guides ” Part of the project . The project has a large number of partners , Include ICC, And completed some preliminary work , For example, verify whether the interpretation of licensing requirements lacks consistency , But there has been little progress in other areas .

Although the inspection engine has been developed , But in “ Express building codes in machine readable form ” There has been little progress . This is related to Fenves stay 20 century 60 The worries of the s are roughly similar , Engineering specifications and other objective 、 Quantitative standards can be implemented in software , But usually the design of this kind of software needs expert guidance and monitoring . Hard coding is often used to translate specifications into software , So the software will become a black box , It is often difficult for people to understand why it comes to such an answer .

today , Research in this field continues , For example, a recent study attempts to implement new zealand building codes in machine readable form , But as far as I know, there is no other implementation .

Part of the difficulty lies in the building code itself , The focus of this work is to translate the specification requirements into clarity 、 Clear rules , But many specifications are not simple .Fenves given 1993 year BOCA An example in the specification ：

If you want to check the above terms through a software , First of all, we must find out “ What is special knowledge and effort ”（special knowledge and effort）, And the definition must be sufficiently detailed , It must cover all the possibilities in the future （ For example, whether a new door handle requires special knowledge and effort ？ A new locking mechanism ？ How to translate it into software evaluable statements , At the same time, there is no need to build a complete model of how the world works ？）

It is not difficult to find similar provisions in building codes , such as 2018 year 《 International housing code 》 The following provisions in ：

Automatic checking of building compliance means , Figure out how to explain the specific meaning to the software . Generally speaking , Creating machine readable building codes means more than just translating the rules themselves , It also requires all the default assumptions about how the world works as reflected in the translation rules . just as Fenves said ,“ Most clauses of the design standards refer to common sense that all human beings should know . Unfortunately , We don't have such a standardized expression of common sense . Everyone naturally knows these common sense .”

These default assumptions will lead to ambiguities in the interpretation of the specification . essentially , The problem we are trying to solve is to eliminate the fuzziness of building codes , Make it easy to interpret , And this problem also makes it difficult for us to solve the problem through software .

against all odds , But I think in solving this problem , We are about to usher in a major turning point . Most previous efforts have been to manually transform specifications into a series of objective rules , But I think in the near future, artificial intelligence system can directly evaluate the existing building codes . Such attempts can be traced back to the last century 90 In the early s , Although there has never been a successful case before . But now language based artificial intelligence models are developing rapidly , We can use OpenAI Of Codex Such a tool explores how to interpret building codes through software .

Create the appropriate building representation

After realizing the representation of building codes in machine readable form , The next step is to represent the building itself in machine readable form .

Most of the early studies of automated specification checking were in hand drawn or two-dimensional CAD Era , In both ways, buildings can be represented by a series of lines , Unfortunately, software cannot understand the meaning . therefore , Early automated specification inspection systems also attempted to create a data structure to represent buildings that the system could understand .

With the last century 90 years BIM The development of , Most automated inspection systems begin with BIM Based on , Study how to create BIM Model to assess compliance . These attempts generally rely on Solibri Or other inspection engines to complete the actual specification evaluation .

There are several challenges in creating a building that can evaluate compliance by machine . One side ,BIM There is no standardized method for model creation , Different designers will use different ways to build BIM Model . This means that the evaluation software must be able to evaluate data models with arbitrary structures , Or force everyone to adopt standardized modeling methods .

Most automatic inspection systems are implemented in the latter way . for example ,SMARTcodes Plan with the United States BIM Combine the development and launch of standards . Singapore is currently trying to automate the process of normative inspection , obviously Corenet-X Also with some very strict BIM Modeling standards work together .

as far as I am concerned ,UpCodes AI Is the only one that does not adopt standardization BIM Model system , This software can evaluate Revit Whether the model conforms to the specification . however , At present, the development of this tool has stalled .

In addition to different modeling styles , Another problem is to create a building model that can be used to check specifications , We have to add a lot of information . As I am 《 BIM、Revit And database dream 》 As mentioned in the article ：

…… Creating a complete building information database requires great efforts . Even medium-sized buildings have hundreds of thousands of components 、 Hundreds of different specifications and Standards , There are countless parameters that need to be specified . Specifying all parameters in advance is a huge job , It also needs long-term maintenance , To ensure that it remains up-to-date . Most design professionals will soon try to save trouble and facilitate maintenance , And omit from the design drawing “ unnecessary ” Details and elements .

These problems are often based on BIM Automatic inspection system . for example ,2009 There was a project in charge of GSA The court creates BIM standard , To ensure that the software can analyze the model , They need to manually mark different spaces , And this work took hundreds of hours , The whole project is in trouble .（ stay AI In the project , The workload of manually marking input data is also very huge , There are similarities with our introduction here .） therefore , Some current automatic checking work , for example SEEBIM Research projects , Are trying to generate such information by computer .

This research continues to advance at a steady rate , But there are not many actual implementation cases . But I think , some time 5～10 During the year AI The model will make great progress in this field .

Output acceptance

The work of creating an automated specification inspection system mainly focuses on problems 1 And 2, That is to create a software representation of specifications and buildings . however , The value of automatic standard inspection system mainly comes from solving the problem of 3 A question , Let stakeholders （ Construction Authority 、 Designers and builders ） Accepting the output of the software is the proof of compliance . Missing this step , All the previous efforts have only automated some repetitive tasks , It has not changed the fundamental process .

At present, the implementation of most automatic standard inspection systems （ In the form of architectural design software ） I didn't even try to solve this problem . for example , Current engineering packages （ Such as RAM） Generally, it can automatically check whether the whole structure of the building meets the specification requirements . Although this undoubtedly saves engineering time , but RAM The printout says “ All inspections have passed ” It will not change the licensing process , After the construction documents are submitted , The program reviewer will still review , And put forward opinions .

This problem is both a technical problem , It is also a political issue . Software needs to generate clear 、 Verifiable output , It also needs to convince jurisdictions and builders who are unwilling to take risks . The most direct way to solve this problem is from top to bottom , Led by the government （ Or at least supported by the government ） Automatic specification inspection items . in fact , Many automated specification inspection projects have taken this kind of promotion , such as SMARTcodes, AutoCodes and CoreNet etc. .

There are other government led automated regulatory inspection systems around the world , stay 2020 At the summit in, Germany 、 European governments such as Norway and Estonia have shown some prototypes . among , The Estonian system seems to be the most advanced , But these systems have not been fully implemented .

Successful projects

above , We introduced a long list of unfinished projects . But there are also some successful projects in this field , such as ResCheck、ComCheck and SolarAPP All are “ In the wild ” The system of , At least they have successfully achieved some of the goals of an automated specification inspection system .

ResCheck and ComCheck It was the U.S. Department of energy in the last century 90 Software developed in the s , Used to verify the energy code compliance of residential and commercial buildings respectively . The two software work in a similar way , The designer inputs the relevant energy parameters into the software （ Through the drop-down menu and text box ）, Then perform heat flow calculation , Finally, compare the specification requirements .ResCheck and ComCheck The difference from other design software is , Its output represents compliance . in fact , Many jurisdictions require that ResCheck or ComCheck As part of the license application , Some will even ResCheck and ComCheck Compliance with local building codes .

SolarAPP It's a similar tool . This is a web-based regulatory compliance tool developed by the National Renewable Energy Laboratory for residential solar installations .SolarAPP Working mode and ResCheck and ComCheck similar , Designers input design parameters into the software through a series of text boxes and drop-down menus , Then the software checks whether it meets the specifications of residential solar devices , Finally, output the compliance report .

but SolarAPP Further more , It can automatically issue building permits in relevant jurisdictions , There is no need for manual plan reviewers to sign it .

SolarAPP Relatively new , It began to popularize only last year . By 2021 year 9 month , America already has 125 Jurisdictions agreed to use the system .NREL Yes 5 A study conducted in four test jurisdictions showed , Today, the average number of people applying for permission is reduced by about one day , The installation and inspection of the project have been accelerated on average 12 God . By 2021 year 12 month ,5 Reviewers in the test jurisdictions saved 2000 More than one hour .

What is the difference between these successful projects and those that failed ？

One side ,ResCheck、ComCheck and SolarAPP The scope of attention is very narrow , Their goal is not to inspect the entire building , Instead, compliance is checked against a very narrow list of specification requirements . therefore , The development speed of such projects is very fast , And it is more convenient to verify whether the output is correct , for example SolarAPP It will output every check it performs , And the output list is very simple , It can be completely rechecked manually .

On the other hand , These projects deliberately avoid some technical problems , These software are relatively simple , It can be operated through text box and drop-down menu . They didn't solve some common problems , For example, transforming specifications into software evaluable rules , Automatically analyze existing BIM Model , I didn't use AI Realize intelligent functions .

let me put it another way , They ignored the problem as much as possible 1 and 2, But focused on solving the problem 3. For example, for SolarApp, Getting stakeholder recognition is the key to the process , Work with jurisdictions to advance development in an iterative manner , Ask for comments and output feedback about the application , Until the output acceptable to everyone is generated .

In my submission , Acceptance is a key obstacle to the development of automated specification inspection system in the future . Maybe in the future 5～10 year （ Or earlier ） AI models can make significant progress in correctly interpreting building codes and building designs, and automatically checking whether buildings comply with codes . But at present, AI models are generally black boxes , No one knows how they get a specific output , Therefore, it will still be a big challenge to use them as a real automatic inspection system .

summary

• If automated specification checking can be popularized , It can greatly reduce the time and effort spent on obtaining construction permits .

• Three major problems need to be solved in automatic specification inspection ： Translate building code requirements into software readable forms , Transform architectural design into a form that software can interpret , And get stakeholders to agree that the output of the software indicates specification compliance .

• At present, we have some software that can automatically perform some specification requirements checks in narrow areas , But it needs the supervision of experts . Most of the work is designing software , But occasionally it involves other disciplines , for example SmartReview Some architectural specification requirements can be checked .

• In order to create a widely used automatic specification inspection system , People have invested a lot of research , And there are many government supported projects , But most of them failed .

• There are few successful cases , Mainly based on specifications within a specific scope , And focus on solving problems 3.

• In the following 5～10 During the year , AI models are likely to solve problems successfully 1 and 2, But the problem 3 Still not resolved , And AI models can lead to problem solving 3 The difficulty of .

• Most of the value of automated specification checking comes from solving problems 3.

Original address ：https://www.infoworld.com/article/3660632/you-re-thinking-about-technical-debt-all-wrong.html

— Recommended reading  —

*AI  Score in Advanced Mathematics  81, Net friend ：AI  The model is inevitable “ Internal volume ”！
* Microsoft banned , Russia is against piracy  Windows  Demand for new products “ Skyrocketing ”！
*“ give up  GitHub , The time has come. ”, Software freedom protection association angrily criticized ！