Researches On The Rate-dependent Low-cycle Fatigue Life Of Polyamide-6 Considering Ratchetting And Its Prediction Models

With the development of supporting industries such as that for synthesis,modification and processing,the polymeric materials have been widely used in different engineering fields.As a major engineering polymeric material,the polyamide-6(PA6)with high specific strength,corrosion resistance,ease for processing,and other excellent properties has been increasingly applied in the fields of automotive industry,mechanical industry,and electronic equipment.In these applications,the load-bearing components made of PA6 are often serviced under a cyclic loading condition,and then their fatigue performance is a critical concern in both design and application.In the past two decades,many experimental and life-prediction researches have been conducted on the fatigue of polymeric materials.However,these experimental studies are more confined to uniaxial fatigue,and the influencing factors of fatigue life have also not been systematically considered yet.The effect of ratchetting(i.e.,the cyclic accumulation of inelastic deformation,occurring in the materials subjected to an asymmetrical stress-controlled cyclic loading)on the fatigue failure of polymeric material has not been well understood.As a result,the existing life-prediction models for polymeric materials have the limitations of reliability and applicability in the engineering applications.A life-prediction model with both high accuracy and broad applicability is a long-term goal of fatigue field.However,since the fatigue mechanism is complicated and far from being fully figured out,the existing life-prediction models generally rely on the simplifications and assumptions with respect to the materials or loading conditions,which limit their accuracy and applicability.In recent years,with the development of data science and the improvement of computational power,the data-driven methods based on machine learning(ML)have been applied to fatigue analysis and provide a new idea for life-prediction modeling.In the latest review papers of life-prediction models,it is prospected that the ML may be the most promising way for developing the complicated life-prediction models.However,the existing studies are still focused on the uniaxial fatigue and rarely extended to multiaxial loading conditions.Besides,these studies are nearly limited to the paradigm of purely data-driven ML,but the fatigue data in practical industrial applications cannot often satisfy the requirement of ML for a "big dataset".Thus,the physics-informed machine learning methods should also be investigated to compensate the insufficient fatigue data.In this study,the whole-life ratchetting and fatigue failure of PA6 are investigated by performing a series of stress-controlled low-cycle fatigue tests,and then a life-prediction model with multiple factors considered is established.Besides,more generalized life-prediction models are established based on the ML and the physics-informed ML life-prediction models are tried from two aspects.The main researches include:(1)The whole-life ratchetting and fatigue failure of PA6 are studied by performing a series of uniaxial and multiaxial stress-controlled low-cycle fatigue tests,in which the effects of stress level,stress rate,and loading path are considered.Besides,the ratchetting-fatigue interaction is discussed,and the competitive relationship between strain hardening and self-heating softening during cyclic deformation,as well as its effect on the fatigue life,is revealed.These experimental results are the basis for establishing the following life-prediction models.(2)In the framework of critical plane approach,a damage parameter is established to fully describe the effects of stress level,ratchetting strain and loading path.Based on the timedependent damage accumulation of polymeric materials,a stress rate function is constructed to characterize the rate dependent fatigue life of PA6.From the above analysis,a semi-empirical life-prediction model is developed for the low-cycle fatigue of polymeric material,and the ratedependent uniaxial and multiaxial fatigue lives of PA6 can be predicted reasonably.(3)A new life-prediction method is proposed based on the ML,in which the characteristics of loading condition is quantified based on the long short-term memory(LSTM)network.It is demonstrated that the proposed method can be well applied in the uniaxial and multiaxial fatigue life predictions of various materials,and most of the predicted lives are located within1.5 times scatter band.This ML-based method breaks through the limitation of conventional life-prediction models,that they generally rely too much on the simplifications and assumptions,and thus only are applied to the specific materials or loading conditions.(4)A new self-attention mechanism-based life-prediction method which is able to characterize the effects of loading history and varying temperature,is proposed based on the ML.In this method,the connections among the material states(i.e.,the loading and ambient conditions)at each time point of loading history are parsed based on the self-attention mechanism.It is verified that the proposed method is well suitable for the following three cases,i.e.,multiaxial fatigue under constant amplitude loading conditions,uniaxial and multiaxial thermo-mechanical fatigue,and multiaxial fatigue under random and variable amplitude loading conditions,and the predicted lives are mostly located within 1.5 times scatter band.(5)According to the life evolution law and fatigue failure mechanism of PA6 obtained in experimental study,the physics-informed ML based life-prediction models are investigated by employing two kinds of specific methods,that is,restriction on the space of updating model parameter,and domain knowledge-based data augmentation.The results show that: the generalization ability of the ML model trained by insufficient data samples can be improved by using the designed physics-informed strategies,and the predicted fatigue lives are consistent with the objective physical laws.