FOCUS Bikes Service Assist

Quick Check

Initial situation

Modern bicycles with carbon frames offer many advantages, especially in terms of lightness and malleability. Faults, defective components or wear are usually noticeable through noise when riding or during servicing. Due to the resonance behavior of carbon frames, the noises are often audible in completely different places and make localization very difficult or even impossible. In addition, there are noises that only occur while riding, e.g. on the saddle or bottom bracket under load. For example, it can happen that something is wrong with the gears but seems to be audible at the front of the stem, so that the owner or mechanic is misguided and looks in the wrong place. This costs a lot of time, parts and patience.

Will it be possible in future to detect faults using just one sensor mounted on (or in future in) the frame, such as a microphone and a smartphone?

Solution idea

Research questions for the Quick Check:

Where should the sensor be installed to ensure optimum detection?
Can one or two error cases be recognized and distinguished?
What is the effort and process involved in training more cases?
How can the algorithms be transferred to other bikes?

Benefit

The training of error cases offers the opportunity to create an application that increases the quality of bicycles, service and safety. The application could also be used as early as the development stage. The Quick Check provides information on whether and how the opportunities could be used to offer previously unprogrammable assistance for different groups, possibly even across manufacturer boundaries. It is even conceivable that such an application could contribute to the sustainability of bicycles if it were also possible to detect wear without having to install additional sensors on the parts, so that parts such as brake pads or tires could be replaced at the right time.

The project also provides information on user interaction with AI and safety on bicycles and in road traffic.

Implementation

Two microphones were attached to the frame of the test bike. One microphone directly on the frame and another on the steering wheel. The microphone recordings were analyzed to assess the potential and feasibility of such a solution. Various faults and problems were defined for this purpose. The aim was to detect a loose brake caliper, for example. The recordings can be analyzed in a variety of ways and give an idea of the potential that can be tapped by an intelligent, adaptive system.

Display of an audio recording of several switching operations

Frequency spectrum over the entire recording

Display of audio features with the help of an MEL spectrogram

MEL spectrogram with Short-Time Fourier Transformation (STFT)

Difference between with or without motor support

The loose brake caliper causes a distinct noise when actuated

In addition to the presumed detection of faults, it is also possible to detect whether gear changes have been made at the right moment. If the rider shifts under load, i.e. with an active motor, a lot of force is transferred to the chain. This leads to increased wear of the chain and the sprockets, especially with a powerful motor, and in the worst case can lead to breakage.

Based on the audio recordings and the analysis, the following possibilities arise, for example:

Detection of continuous changes to the bicycle, e.g. wear, loss of pressure
Detection of defects e.g. loose brake caliper
Detection of maintenance requirements, e.g. greasing the fork or seat post maintenance
Detection of the current surface and information on the appropriate tire pressure
Recording of switching behavior, maintenance requirements or future automatic readjustment of the circuit