+44 (0) 1954 231 494
Cambridge Ultrasonics
Cambridge, UK
Consultancy service in physics, electronics, maths & ultrasonics

Cambridge Ultrasonics


Finite element model (axial symmetry) in 2D of the transmitter transducer used for testing rivets. Performance was optimized using FE.


Rivet inspection

A major automotive manufacturer wanted an inspection system to test the mechanical integrity of rivet joints used in aluminium panels. Cambridge Ultrasonics developed a novel transducer assembly to inspect such rivets.  A circular array of receivers with a single transmitter in the middle was built. The assembly was built to fit over a rivet-head. The first stage in development was finite element modelling to optimize the transmitter. The received signals were processed using an artificial neural network to decide if the rivet under test was faulty or in good condition. The transducer system, complete with  miniature head amplifiers and multiplexers, was used with a data acquisition system in a PC.

Resonance spectroscopy

Another major automotive manufacturer wanted to explore using ultrasonic resonance spectroscopy for quality assurance testing (QA) of mass-produced parts. Resonance spectroscopy is arguably one of the oldest known QA methods in history - tap the object and listen, if it sounds "metallic" and it rings then it is probably good quality, if the sound is "dead" the object probably has a crack in it. Wheel-tapping of railway trains is an example from the 1800s of resonance spectroscopy.

Cambridge Ultrasonics developed novel equipment to inject ultrasound of a constant frequency and amplitude, maintained for a certain period of time, and a receiver to monitor the response of the sample. By stepping through a range of frequencies a spectrum could be created. The spectrum carried information about the test sample. Interpretation in the frequency-domain is more difficult than in the time-domain. It was found that artificial neural networks worked well with mass-produced parts. The difficulties intrinsic in the method are training the artificial neural network and the time taken to collect the spectrum. It is important that at each test frequency sufficient time is taken to allow steady-state resonance to be reached in the sample - this is when the ultrasound has travelled throughout the sample and absorption of ultrasound becomes the dominant mechanism of loss of ultrasonic energy. Failing to wait until steady-state resonance is reached means that it becomes difficult to achieve repeatable conditions of testing and the spectrum cannot be compared easily with other methods that predict the spectrum, for example a finite element program. However, when testing small metallic parts the time to reach steady state can be less than 1 ms, so a spectrum of 1,000 frequencies can be collected in 1 s, which is usually sufficiently short for mass-production.