Handwritten word recognition with character and inter-characterneural networks

An off-line handwritten word recognition system is described. Images of handwritten words are matched to lexicons of candidate strings. A word image is segmented into primitives. The best match between sequences of unions of primitives and a lexicon string is found using dynamic programming. Neural networks assign match scores between characters and segments. Two particularly unique features are that neural networks assign confidence that pairs of segments are compatible with character confidence assignments and that this confidence is integrated into the dynamic programming. Experimental results are provided on data from the U.S. Postal Service.     

Lipoprotein(a) in android obesity and NIDDM

OBJECTIVE: To assess the level of serum lipoprotein(a) [Lp(a)] in nonobese and obese NIDDM subjects with android body distribution. RESEARCH DESIGN AND METHODS: Serum Lp(a) levels were measured in 30 long-standing NIDDM patients (duration of diabetes 12.5 +/- 3 years, mean +/- SD), with 15 of the patients being obese of android distribution (BMI > 30 kg/m2 and waist-to-hip ratio > 0.8). In addition, there were 15 android obese nondiabetic subjects and 10 healthy subjects serving as the control group. RESULTS: All groups of patients in this study (diabetic, obese, and obese diabetic) showed significantly higher levels of Lp(a) than the healthy control group. Lp(a) concentrations were significantly higher in NIDDM patients with android type of obesity than in nondiabetic androids (24.1 +/- 5.6 vs. 14.8 +/- 2.4 mg/dl, P < 0.001). Significantly greater levels of Lp(a) were found in nonobese subjects with diabetes when compared with obese subjects without diabetes (22.3 +/- 4.1 vs. 14.8 +/- 2.4 mg/dl, P < 0.001). Furthermore, Lp(a) serum concentrations were not dependent on the degree of glycemic control (controlled NIDDM 23.6 +/- 5.0 vs. uncontrolled NIDDM 21.4 +/- 2.7 mg/dl, NS), but were much greater in subjects with diabetes complicated by vascular disease (complicated 26.3 +/- 5.0 vs. uncomplicated 20.5 +/- 2.7 mg/dl, P < 0.001). No correlation was found between Lp(a) and other lipid parameters in this study. CONCLUSIONS: Lp(a) levels are significantly elevated in both android-obese and nonobese NIDDM patients regardless of the degree of glycemic control. Lp(a) is an independent risk factor showing greater elevations in those subjects complicated with diabetic vascular diseases.     

Dependence on the motif YIPP for the physical association of Jak2 kinase with the intracellular carboxyl tail of the angiotensin II AT1 receptor

Angiotensin II is the effector molecule of the renin-angiotensin system. Virtually all of its biochemical actions are mediated through a single class of cell-surface receptors called AT1. These receptors contain the structural features of the seven-transmembrane, G-protein-coupled receptor superfamily. Angiotensin II, acting through the AT1 receptor, also stimulates the Jak/STAT pathway by inducing ligand-dependent Jak2 tyrosine phosphorylation and activation. Here, we show that a glutathione S-transferase fusion protein containing the carboxyl-terminal 54 amino acids of the rat AT1A receptor physically binds to Jak2 in an angiotensin II-dependent manner. Deletional analysis, using both in vitro protocols and cell transfection analysis, showed that this association is dependent on the AT1A receptor motif YIPP (amino acids 319-322). The wild-type AT1A receptor can induce Jak2 tyrosine phosphorylation. In contrast, an AT1A receptor lacking the YIPP motif is unable to induce ligand-dependent phosphorylation of Jak2. Competition experiments with synthetic peptides suggest that each of the YIPP amino acids, including tyrosine 319, is important in Jak2 binding to the AT1A receptor. The binding of the AT1A receptor to the intracellular tyrosine kinase Jak2 supports the concept that the seven-transmembrane superfamily of receptors can physically associate with enzymatically active intracellular proteins, creating a signaling complex mechanistically similar to that observed with growth factor and cytokine receptors.     

An automated coding and classification system with supporting database for effective design of manufacturing systems

The philosophy of group technology (GT) is an important concept in the design of flexible manufacturing systems and manufacturing cells. Group technology is a manufacturing philosophy that identifies similar parts and groups them into families. Beside assigning unique codes to these parts, group technology developers intend to take advantage of part similarities during design and manufacturing processes. GT is not the answer to all manufacturing problems, but it is a good management technique with which to standardize efforts and eliminate duplication. Group technology classifies parts by assigning them to different families based on their similarities in: (1) design attributes (physical shape and size), and/or (2) manufacturing attributes (processing sequence). The manufacturing industry today is process focused; departments and sub units are no longer independent but are interdependent. If the product development process is to be optimized, engineering and manufacturing cannot remain independent any more: they must be coordinated. Each sub-system is a critical component within an integrated manufacturing framework. The coding and classification system is the basis of CAPP and the functioning and reliability of CAPP depends on the robustness of the coding system. The proposed coding system is considered superior to the previously proposed coding systems, in that it has the capability to migrate into multiple manufacturing environments. This article presents the design of a coding and classification system and the supporting database for manufacturing processes based on both design and manufacturing attributes of parts. An interface with the spreadsheet will calculate the machine operation costs for various processes. This menu-driven interactive package is implemented using dBASE-IV. Part Family formation is achieved using a KAMCELL package developed in TURBO Pascal.     

Influence of joint detail on the flexural/torsional interaction of thin-walled structures

For a structure formed from two thin-walled open members connected at 90°, torsion applied to one member can result in torsion as well as flexure in the second member, with the magnitude and direction of this torsion as well as flexure in the second member being determined by the type of joint used. Conventional structural analysis would normally assume the presence of only flexure in the second member. The results from a finite element study of structures formed from thin-walled channel sections connected by box, mitre and stiffened mitre joints is presented and an explanation for the behaviour of the different joint types is given. It is shown that for the box joint the warping deformation of the loaded member is the dominant factor in determining the magnitude and direction of the twisting of the second member, whilst this is determined for the stiffened mitre joint primarily by the St Venant rotation deformation of the loaded member. For the unstiffened joint it is shown that the warping and St Venant rotation deformation effects tend to cancel each other out.     

Frequency dependence of the complex dielectric constant of sodium carboxymethyl cellulose

The dielectric constant, ε′, and dielectric loss, ε″, were determined for three solid carboxymethyl cellulose samples having different levels of substitution and different degrees of polymerization over a frequency range of 0.1 – 10 000 kHz at temperatures from 10–60°C. In contrast to the two relaxation processes, γ and β, previously observed in native cotton cellulose, only one relaxation process within a frequency range of 0.1 – 1 kHz was identified. It was found that the dielectric properties do not only depend on the degree of substitution, but also on the weight-average degree of polymerization and uniformity of distribution.     

Effect of the percentage carbon equivalent on the nodule characteristics,density and modulus of elasticity of ductile cast iron

Successful melting and Mg treatment were conducted using a vortex unit which produced spheroidal graphite (SG) Fe ingots having a percentage carbon equivalent (%CE) ranging from 3.782 to 5.240. Microstructure examination revealed graphite nodules embedded in a white matrix (unetched condition). The variation of %CE proved to have a pronounced effect on the nodule characteristics. The nodule count showed a maximum value at %CE of 4.613. For metal-mould ingots it reached a value of 1584 nodules mm^–2, while for sand-mould ingots it reached a value of 970 nodules mm^–2. Nodule size was found to be inversely proportional to the nodule count for both sand- and metal-mould ingots. The nodularity of all ingots was, in general, higher than 95%. The density of SG Fe was found to decrease gradually (at %CE=3.782) from 7242 to 6969 kg m^–3. The modulus of elasticity (E) showed a boat-like curve having minimum values of 189 and 192 GPa for the sand and metal moulds, respectively.     

Impulse breakdown voltages of triggered multistage vacuum gap

Several sealed-off triggered vacuum gaps are connected in series to improve hold-off voltage. The characteristics of impulse breakdown voltage of these series-connected gaps are investigated experimentally. The sum hold-off voltage of series-connected gaps decreases to a unit hold-off voltage when the maximum value of voltage division ratio across the gaps increases to unity. Self-breakdown probability of the series-connected gaps is always higher than that of a single gap under the same conditions. Hence, stage efficiency of the multistage gap decreases with increasing number of stages. Its value is 90 percent with 2-stage gap and 75 percent with 5-stage gap, respectively, under the same voltage division ratio and the same gap length (2.0 mm) in each stage. Triggered breakdown voltage of 2- or 3- stage gap is several hundred volts when all gaps are triggered simultaneously at the peak of the main impulse wave and a working voltage range is nearly 100 percent in this case. The working voltage range decreases with number of stages. Its value is 45 percent with 3-stage gap and 15 percent with 5-stage gap, respectively, when one triggered gap is fired for switching.