Periodic vehicle inspection is essential to maintain road safety and to ensure that vehicles comply with legal requirements throughout their operational life. Since 2012, FSD Fahrzeugsystemdaten GmbH has been responsible according to German legislation as Central Agency for PTI for the development, provision and validation of test specifications and technologies for PTI (periodic technical inspection). Studies have shown that an increase in the required minimum braking efficiency according to legislation significantly contributes to the prevention of accidents and their consequences. Another measure for road safety is to expand brake tests for vehicles that cannot be tested on a brake test bench. For this purpose, the Deceleration Measurement for vehicle classes M1/N1 during road tests is being advanced to a more sophisticated version of the Deceleration Measurement, aiming to achieve a brake performance evaluation comparable to that of the Reference Brake Force Test. Measurement data from test drives, recorded by the PTI adapter equipped with a triaxial accelerometer and rotation rate sensor, are analyzed using deep learning classifiers to detect braking irregularities with high precision. One of the main challenges in developing these classifiers is the provision of sufficient and diversified training data, especially for anomalous or faulty braking systems. Two approaches are employed to overcome this problem: On the one hand, synthetic data generation is used through physical vehicle simulations and modern AI methods, such as GANs or diffusion models. On the other hand, a mechatronic system was developed to record real measurements with wheel-specific, mocked degradations in the braking system. In addition to data generation, the so called Brake Pressure Manipulator will also be used to validate the synthetically generated measurements. The data recorded with the Brake Pressure Manipulator enable a realistic representation of brake anomalies that were previously not sufficiently represented in real training data. The combined datasets from real and synthetic data, along with data recorded using the Brake Pressure Manipulator, enable a considerable expansion of the database, particularly for rare braking anomalies, and will be used to train the classifiers. Initial results from the deep learning classifiers show that they already achieve an accuracy of over 90% on known vehicle models, underscoring the system's performance. The paper provides detailed insights into the development of the deep learning classifiers, their network architectures, and the training methods compared. Additionally, the functionality of the Brake Pressure Manipulator, it’s technical implementation and possible further developments are described. The revised Deceleration Measurement enables the testing of a wider range of vehicles under realistic conditions and the evaluation of dynamic vehicle decelerations. Future work will focus on optimising the classifiers, expanding the database and further developing the test procedures in order to further improve the brake testing technology for PTI and thereby make a further contribution to increasing road safety.