ELSEVIER JSAE Review 16 (1995) 411-414 New Technologies & New Cars Adaptive shift scheduling strategy introducing neural network in automatic transmission Kaoru Kondo a Hitoshi Goka b a Kyoto Transmission Designing, Transmission & Final Drive Engineering Department, Office of Passenger Car Development & Engineering, Mitsubishi Motors Corporation, 1, Tatsumi-cho, Uzumasa Ukyo-ku, Kyoto, 616 Japan b Okazaki Transmission Testing, Transmission & Final Drive Engineering Department, Office of Passenger Car Development & Engineering, Mitsubishi Motors Corporation, 1, Nakashinkiri, Hashime-cho, Okazaki, Aichi 444 Japan Received 14 February 1995 Abstract Conventional automatic transmissions, which have fixed shift patterns, sometimes show inconvenient shifting patterns, especially in uphill and downhill driving. In order to improve the driveability for these road conditions, Mitsubishi developed an adaptive shift logic called "Fuzzy shift" in 1992. Since then, further evolutional shift scheduling strategies has been developed to cover more extensive road conditions. This paper introduces neural network, learning and continuous variable shift patterns control incorporating this strategy. 1. Introduction In most conventional automatic transmissions (A/T), gear selections are determined based on the shift patterns of vehicle speed and engine throttle opening, as shown in Fig. 1. These A/T's sometimes select inconvenient gears, especially in uphill and downhill driving because their shift patterns are set to be mainly suitable for driving on flat roads. For example, when driving uphill, whenever the accelerator is released before a curve, unneccesary upshift- ing is provided, so that when leaving the curve, the accelerator must be pressed until downshift in order to supplement the driving force, and thus smooth driving is disturbed and extra operation of the accelerator is required. In downhill driving, use of a higher gear causes engine brake shortage resulting in frequent foot brake operations to downshift. In these cases, drivers have to use the shift lever or switch to select their desired gears. In 1992, Mitsubishi Motors Corporation introduced adaptive shift scheduling called "Fuzzy shift" to achieve easier driving in these road conditions (Fig. 2). Since then, "Fuzzy shift" has been applied to all passenger vehicles, and has been applauded by domestic and overseas cus- tomers. In order to realize comfortable shift scheduling more suited to individual drivers' intentions in all driving situa- tions, we have developed the adaptive shift scheduling strategy which includes "Neural network" and "Learning control" for downhill driving, and "Continuous variable shift patterns control" for flat road driving and uphill driving. This paper describes this newly-developed strat- egy and functions. 2. Control system Figure 3 is a schematic diagram of this control system. The A/T computer determines the gear based on informa- tion such as the input speed, the output speed, the throttle opening, the foot brake signal, the engine speed, the air flow rate and the steering angle from output signals of sensors or communication with the engine computer. 3. Control method 3.1. Summary The block diagram of the new shift scheduling is shown in Fig. 4. The new shift scheduling consists of three main blocks: the "Neural network" block which calculates engine brake applicability the learning block of engine 0389-4304/95/$09.50 © 1995 Society of Automotive Engineers of Japan, Inc. and Elsevier Science B.V. All rights reserved SSDI 0389-4304(95)00040-2 JSAE9538041 412 K. Kondo, H. Goka /JSAE Review 16 (1995) 411-414 VEHICLE SPEED Fig. 1. Conventional shift pattern. I ROAD GRADIENT I VEHICLE SPEED I BRAKING FORCE I STEERING ANGLE :ONNECTION ENGINE BRAKE APPLICAB L TY INPUT CALCULATION OF RUNNING CONDITION MODE SELECTION DECISION OF SHIFT ENGINE BRAKE MODE STANDARD MODE UPHILL I MOD