As one of the important supporting technologies for the construction of CPS(Cyber-Physical Space), the application of digital twin has appeared in many fields such as aerospace, intelligent manufacturing, smart city, smart medical care, smart education, etc., but some problems in the practice process has also appeared, such as the generalization of the concept of digital twin, which easily leads to misunderstanding; The lack of general and effective methods in concrete implementation, which results in the lack of universal practical cases and difficulty in forming recognized typical cases. To solve these problems, it is necessary to restrict the concept boundary of digital twin and extend the boundary of digital twin method, so as to form a new boundary of digital twin, promote the formation of consensus, increase implementation methods and promote its development better. Conceptually, physical entity is a collection of materials with various properties, which is characterized by complexity, authenticity and immediacy, and can dynamically evolve according to objective laws with the change of external conditions. In the study of mapping physical entities to digital twins, it is easy to expand the scope of concepts, such as the simulation of dynamic evolution process, external conditions, data, and even objects, such as physical entities. As a result, the concept of digital twins will lose its core and it is difficult to reach an agreement. In terms of methods, based on the detailed investigation and analysis of related applications in multi-scenarios, the existing research on digital twin often goes beyond the perception process of physical entities, and directly relies on the models or model building methods in the original professional fields. While this method has made some progress, its limitations have already appeared. First, the existing models are mostly focused on subdivision fields, which cannot be used universally among fields. Secondly, there are few existing models that reflect the characteristics of high fidelity, multi-scale and multi-physical fields of the digital twin model, and even less information flow corresponding to the whole life cycle. Direct application cannot guarantee the effective realization of digital twin; Thirdly, entering the modeling process directly from the height of existing cognition may cause huge costs. For example, the ADT program, a typical case recognized by the US military in the world, has a construction time of ten years and huge investment in manpower and material resources, which further hinders the industry from entering the digital twin practice. In order to meet these challenges, in terms of conceptual boundary, this paper proposes that the digital twin should return to the essence of its digital model, and take the model as the center for effective restraint, so as to promote the understanding of all parties involved in production, education and research to reach an agreement. As for the extension of method boundary, this paper puts forward a method of constructing digital twin model for multi-sensory, that is, according to the general process of human understanding of the physical world-firstly, obtaining perceptual knowledge of features through various perception methods, and then further forming rational knowledge through various cognitive processes-from shallow to deep, from easy to difficult, from simple to complex. Firstly, the digital twin initial model of physical entity is established by multi-sensing methods such as visual perception, auditory perception, tactile perception and dynamic perception, gustatory/taste perception, and combination with control data reflecting the change of conditions, thus focusing on the complexity and authenticity of physical entity at the beginning of the model establishment, fully embodying the characteristics of the digital twin model, and effectively enhancing the practicability and universality of the model; Then, the initial model is gradually matched with the existing cognitive knowledge framework, and the control data returned from the physical entity is used for continuous iteration. In this way, information such as various property changes, real-time/near-real-time reactions, various objective laws and behavioral logics affecting the physical entity under specific external conditions can be gradually added to the digital twin model according to the actual needs of the research field, thus effectively controlling the scale and cost of the model and gradually realizing the information flow in the whole life cycle; Then, the optimized digital twin model is further used in theoretical and practical research, such as simulation, planning, optimization, decision-making, etc., to promote the development of various studies. This paper also reviews and looks forward to the existing technical basis and development prospect of the construction method of digital twin model for multi-sensory.