Can Convolution Neural Networks Be Used for Detection of Gravitational Waves from Precessing Black Hole Systems?

Current searches for gravitational waves from black hole binaries with LIGO and Virgo observatories are limited to using the analytical models computed for the systems with spins of the black holes aligned with the orbital angular momentum of the binary. The observation of black hole binaries with precessing spins (spins misaligned with the orbital angular momentum) could provide unique astrophysical insights into the formation of these sources; thus, it is crucial to design a search scheme to detect compact binaries with precession spins. It has been shown that the detection of compact binaries carrying precessing spins is not achievable with aligned-spins template waveforms. Efforts have been made to construct the template banks to detect the precessing binaries using the matched filtering based detection pipelines. However, in the absence of robust methods and huge computational requirements, the precessing searches, more or less, still remain a challenge. This work introduces a detection scheme for the binary black holes to classify them into aligned and precessing systems using a convolution neural network (\CNN). We treat the detection of the $\GW$ signals from $\BBH$ systems as a classification problem. Our architecture first classifies data as $\GW$ signals originated from binary black holes ($\BBH$) or noise and then it further classifies the detected $\BBH$ systems as precessing or non-precessing (aligned/anti-aligned) systems. Our classifier with an accuracy of $\approx$ 99\% classifies between noise and the signal and with an accuracy of $\approx$ 91\% it classifies the detected signals into aligned and precessing signals. We have also extended our analysis for a multi-detector framework and tested the performance of our designed architecture on $\O1$, $\O2$ and $\O3$ data to identify the detected $\BBH$ events as aligned and precessing events.