Marvel: A Data-Centric Approach for Mapping Deep Learning Operators on Spatial Accelerators

A spatial accelerator’s efficiency depends heavily on both its mapper and cost models to generate optimized mappings for various operators of DNN models. However, existing cost models lack a formal boundary over their input programs (operators) for accurate and tractable cost analysis of the mappings, and this results in adaptability challenges to the cost models for new operators. We consider the recently introduced Maestro Data-Centric (MDC) notation and its analytical cost model to address this challenge because any mapping expressed in the notation is precisely analyzable using the MDC’s cost model. In this article, we characterize the set of input operators and their mappings expressed in the MDC notation by introducing a set of conformability rules . The outcome of these rules is that any loop nest that is perfectly nested with affine tensor subscripts and without conditionals is conformable to the MDC notation. A majority of the primitive operators in deep learning are such loop nests. In addition, our rules enable us to automatically translate a mapping expressed in the loop nest form to MDC notation and use the MDC’s cost model to guide upstream mappers. Our conformability rules over the input operators result in a structured mapping space of the operators, which enables us to introduce a mapper based on our decoupled off-chip/on-chip approach to accelerate mapping space exploration. Our mapper decomposes the original higher-dimensional mapping space of operators into two lower-dimensional off-chip and on-chip subspaces and then optimizes the off-chip subspace followed by the on-chip subspace. We implemented our overall approach in a tool called Marvel , and a benefit of our approach is that it applies to any operator conformable with the MDC notation. We evaluated Marvel over major DNN operators and compared it with past optimizers.

Low-Cost Model for Battlefield Wound and Hemorrhage Training

ABSTRACT Introduction Exsanguination is the leading cause of preventable death on the battlefield and in austere environments. Multiple courses have been developed to save lives by stopping hemorrhage. Training for this requires simulation models; however, many models are expensive, preventing the further expansion of this life-saving training. We present a low-cost model for hemorrhage training and realistic moulage based on simple medical supplies and grocery store meats. Materials and Methods Wound packing training was completed by use of a block of pork shoulder roast with an incision simulating a wound and IV tubing connected to a syringe with fake blood. Hemostasis was obtained with proper wound packing by the student, causing the bleeding to be tamponaded. Wound moulage utilized remaining supplies of pork roast being attached to patient actors or mannequins and adorned with fake blood creating wounds with the appearance and feel of real tissues. Results Tactical Combat Casualty Care (TCCC) training was completed at a small military medical facility with a start-up cost of less than $70 and a single course as cheap as $15. These methods have been utilized to establish other TCCC training centers while keeping costs low. Conclusions We present low-cost models for simulating massive hemorrhage for wound packing with pork roast and realistic moulage. These methods can be utilized for other hemorrhage training courses such as TCCC, Advanced Wilderness Life Support, and Stop the Bleed.

Technical–Economic Feasibility Analysis of Subsea Shuttle Tanker

This paper presents the technical and economic feasibility analysis of the subsea shuttle tanker (SST). The SST is proposed as an alternative to subsea pipelines and surface tankers with the primary purpose of transporting CO2 autonomously underwater from onshore facilities to subsea wells for direct injection at marginal subsea fields. In contrast to highly weather-dependent surface tanker operations, the SST can operate in any condition underwater. The technical–economic analysis is performed in two steps. First, the SST’s technical feasibility is evaluated by investigating designs with lower and higher capacities. The purpose is to observe the appearance of technical limits (if present) when the SST is scaled down or up in size. Second, an economic analysis is performed using the well-reviewed cost models from the publicly available Zero Emissions Platform (ZEP) and Maritime Un-manned Navigation through Intelligence in Networks (MUNIN) D9.3 reports. The scenarios considered are CO2 transport volumes of 1 to 20 million tons per annum (mtpa) with transport distances of 180 km to 1500 km in which the cost per ton of CO2 is compared between offshore pipelines, crewed/autonomous tanker ships, and SST. The results show that SSTs with cargo capacities 10,569 m3, 23,239 m3, and 40,730 m3 are technically feasible. Furthermore, the SSTs are competitive for short and intermediate distances of 180–750 km and smaller CO2 volumes of 1–2.5 mtpa. Lastly, it is mentioned that the SST design used the DNVGL Rules for Classification for Naval Vessels, Part 4 Sub-surface ships, Chapter 1 Submarine, DNVGL-RU-NAVAL-Pt4Ch1, which is primarily catered towards military submarine design. It is expected that a dedicated structural design code that is optimized for the SST would reduce the structural weight and corresponding capital expenditure (CAPEX).

Peak Shaving and Frequency Regulation Coordinated Output Optimization Based on Improving Economy of Energy Storage

In this paper, a peak shaving and frequency regulation coordinated output strategy based on the existing energy storage is proposed to improve the economic problem of energy storage development and increase the economic benefits of energy storage in industrial parks. In the proposed strategy, the profit and cost models of peak shaving and frequency regulation are first established. Second, the benefits brought by the output of energy storage, degradation cost and operation and maintenance costs are considered to establish an economic optimization model, which is used to realize the division of peak shaving and frequency regulation capacity of energy storage based on peak shaving and frequency regulation output optimization. Finally, the intra-day model predictive control method is employed for rolling optimization. An intra-day peak shaving and frequency regulation coordinated output optimization strategy of energy storage is proposed. Through the example simulation, the experiment results show that the electricity cost of the whole day is reduced by 10.96% by using the coordinated output strategy of peak shaving and frequency regulation. The obtained further comparative analysis results and the life cycle economic analysis show that the profit brought by the proposed coordinated output optimization strategy is greater than that for separate peak shaving or frequency modulation of energy storage under the same capacity.

Toward cost-efficient tolerancing of 3D-printed parts: a novel methodology for the development of tolerance-cost models for fused layer modeling

Tolerance allocation methods significantly contribute to the qualification of Additive Manufacturing (AM) for (small-)series production ensuring high performance and efficiency. However, their usage prerequisites the availability of quantitative, reliable information on the impact of the assigned tolerances on the resulting manufacturing costs. The given article proposes a novel methodology for the systematic development of tolerance-cost curves for a cost-efficient tolerancing of 3D-printed parts. The proposed structured workflow aims at serving as a general guideline for both researchers and practitioners, while the exemplarily chosen perspective from Fused Layer Modeling (FLM) illustrates its adaption to a specific AM technology. The indirect, non-apparent interrelations between tolerances and resulting costs are modelled with the aid of an activity-based cost model, whereas the individual costs elements are mapped as function of the values for the machine-specific process parameters for AM, e.g., layer height or printing speed, which are required to achieve the assigned design tolerances. The total procedure covers all relevant steps, viz. the identification and quantification of the single cost items, the design of benchmark artifacts, adapted to given manufacturing and measuring techniques, the empirical determination of data on cost and geometrical accuracy by design of experiments and tolerance-cost curves. Its exemplary application to an academic use case shows its general applicability and benefits, but also its current limitations.

Development of a Cost Model for Vertical Milling Machines to Assess Impact of Lightweighting

Lightweighting is a design strategy to reduce energy consumption through the reduction of mass of a product. Lightweighting can be applied to machine tools to reduce the amount of energy consumed during the use phase. Thus, the energy cost of machine operation will be reduced. One might also hypothesize that since a lighter-weight machine tool requires less material to build, the cost to produce such a machine will be less. However, it may also be the case that lightweighting a machine tool increases its complexity, which will likely drive up the cost to manufacture the machine. To explore the cost drivers associated with building a machine tool, data on the features associated with a wide variety of vertical milling machine tools are collected. Then, empirical cost models are fit to this data. The results from the cost models show that the machine tool mass is a significant cost driver; other key drivers are the number of axes and spindle power. The models are used to predict the cost benefits of lightweighting in terms of mass, which are compared to potential increased manufacturing costs associated with complexities introduced due to lightweighting.

The Impact of Weather and Slope Conditions on the Productivity, Cost, and GHG Emissions of a Ground-Based Harvesting Operation in Mountain Hardwoods

Mountainous hardwood mixed stands offer challenges to timber harvesting operations in practice, including a harsh climate, variable topography, steep terrain, and large-sized timbers. This paper aims to develop productivity and cost models for a mountain-ground-based harvesting operation across the terrain (e.g., slope conditions), stand (e.g., tree volume) environmental (e.g., weather), and yard (e.g., winching distance) variables and to assess GHG emissions related to the equipment in use. This development was implemented in a timber harvesting practice under single-tree selection in mountainous forests of Iran where a motor-manual chainsaw is used for felling and a rubber-tired cable skidder is used for log extraction. The average delay-free productivity was 4.55 m3 for felling and 14.73 m3 h−1for skidding. Lower production costs and higher productivity rates were observed over the gentle slopes and in sunny conditions. The average production costs ranged between USD 4.27m−3 for felling and USD 5.35m−3 for skidding. The average emissions ranged between 0.96 kg m−3 for felling and 7.06 kg m−3 for skidding in snowy conditions over steep slopes. The study’s results confirm avoiding harvesting operations on steep slopes (greater than 35%) and in extreme weather conditions to obtain higher work efficiency and to minimize adverse effects of machinery on forest ecosystems. The results should be of use to harvest managers and forest planners considering the application of ground-based harvesting operations using a semi-mechanized system on a range of operating conditions in mountain hardwood stands.