A Proof-of-Concept Convolutional Neural Network for Estimating the Total Mass of Binary Black Hole Systems from their Gravitational Waves

Gravitational wave detections provide information about the physical properties of their source binary system, such as the total mass of the system. Typically, these parameters are extracted by comparing observed gravitational wave signals with waveform models. This study presents a proof-of-concept investigation of how a convolutional neural network (CNN) can estimate the total mass of a binary black hole system from its gravitational wave waveform. Simulated gravitational waves are generated over a range of total masses using a gravitational wave waveform model. The simulated waves are then used as inputs to a one-dimensional CNN. The proof-of-concept implements fixed values for many parameters for both the binary system and the detector. This is done to isolate the relationship between the waveform and the total mass. A preliminary CNN implementation was developed and evaluated using simulated gravitational wave waveforms. The preliminary implementation generated 1,000 simulated waveforms with total masses linearly spaced from 50 to 600 M☉ and used them as inputs to a one-dimensional CNN. Many source and detector parameters were held fixed to isolate the relationship between waveform morphology and total mass. On the held-out test set, the implementation produced a mean absolute error (MAE) of 33.94 M☉ and a root mean squared error (RMSE) of 45.13 M☉. Because the implementation was not independently validated and used a restricted set of simulated source parameters, its results are presented as a proof-of-concept rather than as evidence of a validated predictive model. The study also discusses its limitations, including the use of simulated data, restricted source parameters, and simplified detector conditions, as well as potential extensions applicable to more realistic gravitational wave waveforms.