Complexity Study On Several Chemical Engineering Systems

This work aims to study some chemical engineering systems based on the complex sys-tems. We hope to discover some universal laws, which could be applied in the real chemical engineering systems to explain some emerged phenomenons. In this work, three real systems, that are the foaming polymer, atmospheric turbulence in the wind tunnel test and the drug-target system, are taken as the examples to research the emerged complexity laws. The complexity theory and methods such as the statistical physical, multifractal, complex network are adopted to investigate the above systems.In Chapter 1, we firstly give a brief introduction of the complex system and the complexity science respectively, summarize the complex system characteristics such as the emergence and self-similar, and present the role of the complex network in the complexity research. Than we illustrate the complexity in the chemical engineering systems mainly based on the multi-scale effect. Finally, we outline the main content and the thesis structure.In Chapter 2, the two-dimensional multifractal detrended fluctuation analysis (2D-MF-DFA) is applied to reveal the multifractal properties of the fracture surfaces of foamed polypropy-lene/low density polyethylene (PP/LDPE) blends at different experiment conditions. Nice power-law scaling relationship between the detrended fluctuation Fq and the scale s is ob-served for different orders q and the scaling exponent h(q) is found to be a nonlinear function of q, confirming the presence of multifractality in the fracture surfaces. The multifractal spec-tra f(a) are obtained numerically through Legendre transform. Two fundamental quantities related to the multifractal spectrum (the multifractal spectrum width Aαand the difference of the fractal dimension△f) change regularly with the experiment conditions. Since the amount of the PP spherulite (hard foam range) reduced with augmenting LDPE accretion in the mix-ture samples, the resultant cell densities of foamed blends are increased, which indicates that the fracture surfaces become more and more irregular and complex. In the multifractal formal-ism, the result is that Aa increases and△f decreases with the increase of LDPE fraction. For higher temperatures cause larger foamed regions with more cells and larger cell size, the influ-ence of the multifractal feature with increasing temperature is similar to the increase of LDPE content. In the foaming precess, through the uneven cell nuclear and the self-organization in-teraction of cell growth, the self-similar fracture surface is emerged. The multifractal spectrum can serve as "complexity measures" of the fracture surfaces of foamed polymer to illustrate the morphological features.In Chapter 3, we investigate the statistical properties of the extreme events intervals r between successive energy dissipation rates above a certain threshold Q in the atmospheric turbulence in the wind tunnel test, aiming at unveiling the laws governing the occurrence of extreme events. We find that the distribution function Pq(r) scales with the mean return interval RQ as PQ(r)=RQ-1f(r/RQ) for RQ∈[50,500], where the scaling function f(x) has two power-law regimes. The scaling behavior is statistically validated by the Cramer-von Mises criterion. According to the scaling behavior, we calculate the probability WQ(△t,t) which indicates the reoccurring probability of the extreme event after a short timeδt, and the empirical curve conforms the validation of the prediction. The conditional distribution PQ(r|r0) and the detrended fluctuation analysis (DFA) result indicate that there are short memory and long-term memory in the interval series respectively. The Hurst index H exhibits exponential decay with RQ, and for very extreme events with infinity RQ, H can be predicted as H∞=0.639. Finally, we find the multifractal property in the return intervals adopted the MF-DFA method. Through the non-multifractal features of the shuffled series, it is found that the fat tail distribution and the long-term memory may be the origin of the multifractal in the interval series.In Chapter 4, we investigate the turbulence signals with the complex network theory and method by converting time series into networks. We first map the turbulence signals into net-work with the visibility graph method. The power-law degree distribution of the visibility graph shows the self-similar of the turbulence signals and the exponential decay of the box numbers versus the box size in the the edge-covering box-counting method indicates that the visibility graph is not fractal. The largest betweeness centralities spanning tree is adopted to analysis the allometric scaling, and networks mapped from the signals in different segments present similar allometric scaling properties with the allometric scaling exponentsη=1.163±0.005. Than we construct the nearest network with the turbulence signals. Through the motif analysis, we find that the turbulence signals with different time scale exhibit different motif-rank patterns, which confirms the multi-scale effect in the turbulence. The motif-rank patterns of the nearest network transform regularly with the change of time scale. With the superfamily phenomenon of time series characterized by the motif-rank pattern, the turbulence scale could be classified into different categories, but the switching values of scales cannot be related to the well-known turbulence scales such as the the dissipation subrange and the inertial subrange. According to the simulation result of the FBM and MRW, we find that the origin of the different motif-rank patterns for different turbulence scales should be determined by the DFA exponent.In Chapter 5, we investigate the recommendation system and adopt recommendation algo-rithm in the drug-target system to predict new drug-target interactions. Firstly, we research the personalized recommendation algorithm to get more accurate and diversiform recommendation result. There are two basic network structure based recommendation methods:the ProbS al-gorithm which based on the mass diffusion and the HeatS algorithm which based on the heat conduction. We investigate the effect of heterogeneous initial resource configurations over the ProbS+HeatS hybrid recommendation algorithm. We test the performance of the algorithm that HeatS+Probs with heterogeneous initial configuration (HPIC) on the MovieLens and Netflix data sets. Through the analysis of the ranking score, precision and recall, we conclude that the HPIC method would recommend more accurate result than the HeatS+ProbS algorithm. The di- versity is also improved with the HPIC method according to the result of the intra-user diversity Dintra and the inter-user diversity Dinter of the recommendation list. Than, the recommendation algorithm is introduced into the drug-target system to provide more likely drug-target interac-tions for the biological test using the recommendation algorithm. In this work, we present three methods to predict new drug-target interactions:drug-based similarity inference (DBS), target-based similarity inference (TBS) and network-based inference (NBI). The performances of the algorithm on the benchmark data set indicate that the recommendation result obtained from the NBI method is more accurate than the other methods, also outperform the results reported in the literatures. Three target protein dipeptidyl peptidase-IV (DPP-IV), estrogen receptorsα(ERa) and estrogen receptorsβ(ERβ) are selected to validate the predictions experimentally. From 40 drugs which located at the topside of the recommendation list of the corresponding target protein,7 new drug-target interactions are confirmed validation via the in vitro assays. With the biological analysis, we find that the drugs exhibit novel poly pharmacological features on the corresponding target, and could be applied to treat the relevant disease. The results indicated that recommendation system could be powerful tools in prediction of drug-target interactions and drug repositioning.