Translations of linear temporal logic to Büchi automata are key to explicit model checking, with the Tableau-based algorithm commonly used for the translation. This paper describes an improved Tableau-based algorithm which converts linear temporal logic to transition-based Büchi automata. The satisfaction information of the U-formula for the states and transitions is used to judge whether the automata is acceptable using only one acceptance condition set, rather than multiple sets. Thus, the algorithm gives an on-the-fly degeneralization mechanism that can expand the state nodes while detecting their validity and removing invalid nodes. The algorithm combines equivalent states and transitions and avoids simplifications after forming the final automata. Comparisons with other conversion tools show that this algorithm runs faster with less states and transitions in the automaton.