This study aims to understand the effect of polymer chain breaking and healing on the strain field ahead of the crack tip. Polyampholyte (PA) hydrogels are used as the system to learn about chain breaking and healing for transient bonds. Polydimethylsiloxane (PDMS) is used as the system to learn about the time-dependent damage of permanent bonds. Firstly, for rate-dependent material, the constitutive model plays an important role in the analysis of mechanical properties, and model parameter fitting is the key to good development of the constitutive model. To get optimal model parameters fast and automatically, we propose an efficient method to determine the model parameters using machine learning (ML) algorithms together with singular value decomposition (SVD). SVD compresses training data and provides outputs for the ML algorithm. The trained ML algorithm rapidly computes the material responses for a large set of material parameters. We test our method by performing model fitting for PVA model (4 parameters), PA model with chemical crosslinks (9 parameters) and PA model without chemical crosslinks (13 parameters). While directly evaluating the constitutive model millions of times takes hundreds of hours, the ML prediction takes less than one hour for the same fitting effect. Secondly, we study the strain-dependent chain breaking and healing for transient bonds in PA gels. PA gels are nonlinear viscoelastic, with time-dependent behavior controlled by the breaking and reforming of ionic bonds in the dynamic network. Relaxation experiments are performed on single edge notch tension (SENT) and T shape specimens consisting of different variations of polyampholyte (PA) hydrogels. In contrast to the linear viscoelastic theory, faster relaxation speed in the higher strain region is observed in both SENT and T-shape samples, which demonstrates the strain-dependent chain dynamics in PA gels. We further find the load transfer between permanent and dynamic networks of different strengths. This load transfer mechanism is connected to viscoelastic behavior. All these experimental results are explained by a nonlinear viscoelastic model. Thirdly, we study the time-dependent chain breaking of permanent networks by studying the delayed fracture of PDMS. Here we study the interaction between polymer chain damage and the elastic field experimentally using different specimens and crack geometries with blunt and sharp cracks. We find that stable slow crack growth can occur in sharp crack samples within a wide range of applied load, while catastrophic fracture can happen in blunt crack samples after hours of holding. Our experiments demonstrate a universal relation between crack growth rate and applied energy release rate in sharp crack samples. A model coupling nonlinear elastic fracture mechanics and rate-dependent bond scission is proposed to explain and describe this relationship.