CROSS Safety Report
Further example of incorrect finite element modelling
Following on from a recent CROSS report, further concerns in respect of the sufficiency of finite element modelling are reported, in particular, the modelling of masonry walls as part of a reinforced concrete framed building.
Key Learning Outcomes
For civil and structural design engineers:
- Model masonry walls as loads only, other than in exceptional cases
- Use the simplest model that is reasonably practicable
- Remember that analytical methods are approximations of true performance
- Be aware of time and temperature related changes such as creep, shrinkage and expansion
- Attributes such as ductility are fundamentally important for the safety of structures
- Ensure that those using specialist software programs understand the materials being modelled
- It is good practice to carry out sense checks and validate all analysis and design outputs
Find out more about the Full Report
Our secure and confidential safety reporting system gives professionals the opportunity to share their experiences to help others. If you would like to know more, please visit the reporting to CROSS-AUS page.
CROSS report 1073 Concern over modelling of concrete frame building for construction stage reminded a reporter of instances where different but in some ways similar, gross errors came to their attention. One of these occurred when they were checking a full-building finite element model, but they have also seen errors elsewhere. This particular problem was picked up in checking, dealt with, and all was well. However, in talking to colleagues and software developers about their concerns, the reporter has become less confident that such issues will be picked up because there seems very little feedback and training.
The reporter goes on to say the concerning model included loadbearing and non-loadbearing walls which were intended to be constructed in masonry. The primary part of the structure was of in-situ reinforced concrete columns and slabs including a transfer slab. There were some beams but in the main, it was flat slab construction. The walls had been modelled with full attachment to the slabs with no releases other than a moment release for transverse bending. The reporter contends that this form of model will give grossly incorrect results in most circumstances and is a worrying situation. There are at least two effects at play which, according to the reporter's ‘asking around’ do not seem well understood by less experienced staff and some more experienced designers.
Firstly, walls will arch, and they are usually stiff enough to arch even when the masonry is much weaker and less stiff than the frame. Indeed, this can sometimes be seen in practice. The result is that the beam or slab that the engineer thinks is supporting the wall, is not receiving that load, especially near the middle of the span, so slab or beam bending moments may be significantly incorrect.
The effect is to introduce non-conservative errors, possibly gross errors.
Secondly, a full connection (where in-plane shear can be transferred) will inevitably cause the wall and slab or beam to behave compositely. If the wall is of substantial height, say a storey height, it will have appreciable stiffness. The effect is to introduce non-conservative errors, possibly gross errors. Shear outputs will be affected too, however, it depends on how the wall is attached to columns, other walls, or other elements. It might be expected that most of the shear would get back to the ends, but that would not be the case if the wall has been inadvertently attached to the column, remarks the reporter.
The reporter contends that design errors will propagate regardless of whether the designer designs manually, or allows the software to generate the reinforcement from the model.
The reporter has experimented with various software packages to try to evaluate if there are any easily implemented ‘workarounds’. The necessary releases can be made, but this is not always a straightforward task. It may also be complicated by how the software makes these releases. For example, some software can release the longitudinal shear along an edge, but still have the end/corners attached. Getting all this right is possible, but it is fiddly and, the reporter fears, beyond many who are routinely creating these models.
Another thing that might tempt the user, is to ‘soften’ the wall material. However, this is of limited use, because the wall is likely so stiff it would have limited effect. A further idea might be to directly adjust the element stiffness matrix to make it incapable of interacting in this way. Doing so however could result in elements that are badly conditioned, and which may require much study to validate whether they are behaving and interacting as expected.
Doing so could result in elements that are badly conditioned
The reporter would ideally like to see a special wall element that is a compression-only element and unable to interact in the sense of longitudinal shear at the edges and corners. The reporter has only seen one software system that might implement something like this. Wall elements can be arranged to form vertical strips – these act like a series of columns that are effectively in contact to form a wall. They might gather transverse load, such as wind, by using dummy load collection elements which are quite common.
The reporter would suggest that if a designer is not confident that they can predict behaviour or interpret the results, they stick to modelling masonry walls as loads only so that unwanted interactions cannot occur. This is inconvenient but tends to be on the safe side.
An objection might arise if walls do indeed arch. Although arching relies on things that may or may not be under the control of the designer. What happens if, for example, someone later puts a door in? What happens if a wind post is put in?
The reporter goes on to say that, nearly all such modelling is done with largely blind faith in linear elastic models. If different forms of elements exist in the model, construction does not occur instantaneously and simultaneously, and there are materials with quite different stress-strain behaviours and amounts of ductility, getting anything like a realistic result probably gets less likely with increasing complexity (and arguably software capability), especially if one injects the question of user expertise. So, for most projects, the reporter would support the principle of using the simplest model that is reasonably practicable, and that computer modelling is a force for good, in the sense that, properly implemented, it promotes economy and reduces error. However, if used incorrectly the scope for error can seem disproportionate.
using the simplest model that is reasonably practicable
The reporter concludes that these issues concern mainly less experienced staff, but fears that those people are progressing rapidly to positions where they are in a supervisory role, without having sufficient training. Additionally, feedback within the industry does not seem reliable. The reporter knows that there are companies in the UK (and possibly elsewhere) that try to use less expensive staff for this kind of work, maybe in the belief that it is sufficient to be able to ‘drive’ a computer and the software takes care of the rest. The reporter believes, however, that these roles deserve a higher profile than that and more educational input is required.
Submit a report
Your report will make a difference. It will help to create positive change and improve safety.
Our secure and confidential safety reporting system gives professionals the opportunity to share their experiences to help others.
No feedback has yet been published for this page.
Expert Panel Comments
An Expert Panel comment on the reports we receive. They use their experience to help you understand what can be learned from the reports. If you would like to know more, please visit the CROSS-AUS Expert Panel page.
The reporter raises a number of very valid points and emphasises that designers should understand the medium they are working in and that analytical methods are approximations of true performance. These approximations and their influence need to be well understood. CROSS report Unconservative design of flat slab due to software modelling issues reviewed similar issues where masonry was modelled as part of the supporting structure – this report also noted the improper use of (or over reliance on) computer modelling with potential for results to be divorced from reality. As noted by the reporter, CROSS report Concern over modelling of concrete frame building for construction stage explored a number of further issues related to modelling including the importance of verification processes.
The idea that engineers can do all analyses and design in a single model should be challenged. Software marketers who may not have sufficient oversight from those who actually understand how the programs work may be tempted to over-sell what can be modelled. Designers however who have been taught by engineers are more likely to understand why modelling masonry as part of a steel or concrete structure is very difficult.
modelling masonry as part of a steel or concrete structure is very difficult
Finite element analysis (FEA) in its standard form only addresses linear elastic materials. There are tricks that some software developers employ to allow the use of tension or compression-only 1D elements in such an analysis, by negating their action when they are subject to axial forces in the "other" direction. These options are not suitable for compression or tension-only 2D elements, so in these instances advanced nonlinear explicit solvers are needed. These programs are usually beyond the budgets of all but a few specialist engineering firms.
Finite element analysis requires engineers to understand all the materials being modelled, however, masonry is perhaps not addressed in many engineers’ academic journeys and therefore many designers do not appreciate how modelling should take account of the engineering properties and use of masonry on site. The reporter is right to be concerned about the education basis of modelling. Arching of masonry, for example, must be carefully considered, especially when the masonry is built upon a flexible support. When each course is laid, the mortar can carry a little compression (though not much, hence the limits on how high masonry can be built in a day) and less shear. Hence arching is unlikely to occur for the wall's self-weight unless the supporting structure is propped during the wall's construction. What is then likely to happen is that subsequent loading on an under-stiffness supporting slab or beam will try to pull it away from the masonry above, causing it (the masonry) to attempt to arch and certainly to crack.
Squashing masonry in as shear panels will also certainly invalidate FEA stress predictions
We also have changes of time and temperature to contend with. In this example, we have an interaction between masonry and concrete both of which exhibit time changes by creep and shrinkage and expansion and in the case of these two materials to differing degrees and possibly in opposite directions. Squashing masonry in as shear panels and not allowing soft top packs (for vertical frame shrinkage) has caused problems but will also certainly invalidate any FEA stress predictions. It must be remembered that the detailing of the joints between the concrete and masonry will allow the designer's intent to be achieved. If, however, the designer is not clear about how the structure behaves, how are they able to decide the details needed for a safe structure?
Finite element analysis of masonry can be used, with great care, to gain an insight into how forces can flow through masonry, but should not be relied on, for the design or detailed understanding of it. Except in exceptional cases, masonry should only be represented in steel and concrete FEA models as loads, with independent analysis and design checks being made on the masonry panels themselves.
Ductility is fundamentally important
Finally, we should not forget that originally all analysis was linear elastic but alarm spread when real stresses were measured in frames and it was realised there was little correlation. Out of that came the concept that design ought to be based on ultimate loads. Design then largely evolved into a procedure meaning 'do it this way and by long experience, we know the frame will be good enough'. Although this process might break down with more complicated buildings, the concept of ultimate design is quite adequate provided the structure is, as intimated by the reporter, ductile enough. However, that quality of ductility is elusive and not directly calculable and is achieved largely by detailing which is not always under the main designer's control. Some may consider it fundamentally wrong to excessively manipulate analyses to try and control stresses on the premise that stress governs everything when in real life we know that is not the case. Other attributes such as ductility are fundamentally important for the safety of structures.
Readers might find the following related reports of interest:
Report 1139 Connection fixity considerations for steel frame modelling published in 2022. This report concerned mismatches that are created when connection modelling assumptions are not followed through into the design of steel structures – the mismatches potentially leading to unsafe structures.
Report 1144 Incorrect modelling of a cantilever, also published in 2022, concerned a structure modelling error, that could have led to safety concerns.