Artificial intelligence for determining the slip-resistance class of laid elastic floor covering
IGF 21357 N
About 45 % of the tripping, slipping and falling accidents in Germany occur while walking on flat surfaces.
In the event of an accident, it is often claimed that improper cleaning and maintenance was the cause of the accident, with all claims for damages often being passed directly on to the building cleaner.
The evaluation of the slip-resistant properties of floor coverings is carried out by means of an internationally unique classification system in slip-resistance classes (so-called R-classes).
At present, this classification can only be carried out by means of a stationary test procedure, the “inclined plane” walking procedure, in the laboratory.
Here, the floor covering to be tested is walked on an inclined plane by a test person with a special test shoe while successively increasing the angle of inclination until slipping occurs.
The angle of acceptance determined in this way is correlated with a slip resistance class according to DIN 51130. The disadvantage of this method is that a non-destructive examination of installed floor coverings on site is not possible.
Mobile test methods, such as those using sliding friction measuring devices, only capture partial aspects of the complex system of slipping and neglect the dynamic step kinematics of humans.
Step simulators attempt to reproduce this, but are only used on a laboratory scale due to their size and complexity. In addition, these methods are time-consuming and labour-intensive.
The aim of this project was therefore to develop an automated, non-destructive method for on-site determination of the slip resistance class of resilient floor coverings.
The measurement procedure was realised using artificial intelligence, which predicts the R-class of resilient floor coverings based on measurement curve profiles of the characteristic parameters elasto-plasticity, friction and surface roughness, taking into account mechanical abrasion.
Under the applied laboratory test conditions, temperature and relative humidity showed no significant influence on the R-class rating.
The artificial neural network Multilayer Perceptron (MLP) used to predict the R-class of resilient floor coverings was trained using the Python programming language and open-source development tools.
The validation results of 23 different floor coverings showed that the AI is able to accurately predict the acceptance angles of different floor coverings with a prediction range of about 3°.
The KIMM measuring unit (functional model) developed consists of a housing that is placed in a fixed position on the floor covering, in which sensors for measuring the characteristics are integrated. A microprocessor unit controls the measurements and records the results.
The sensors installed were a force sensor (stationary measurement, vertical movement in relation to the floor covering) to determine the elasto-plasticity (measurement of compressive force and release), a laser distance sensor to measure the roughness (horizontal movement in relation to the floor covering) and a force sensor (horizontal movement in relation to the floor covering) to measure the sliding friction force.
After optimising the mobile measuring unit KIMM based on the achieved results, it will be able to generate meaningful and valid data. This was confirmed by comparing the predicted and actual acceptance values of the floor coverings.
The research report is available on request from FRT.